Technology

Imagine you’re a company. You’ve got sensitive data you need kept safe, and you’ve got a cloud service provider that’s protecting it. But do you know how far that protection goes, and at what point it’s your turn to take over?

The sheer number of data breaches due to human error, specifically database misconfiguration, suggests that many don’t know that they, not the provider, are responsible for securing all the data properly (or are unsure how to do it).

In this article, we’ll look into the shared responsibility model: what it is, why it’s relevant, its benefits, its limits, and what happens when companies disregard it.

What is the Shared Responsibility Model?

The Shared Responsibility Model is a security and compliance framework that defines the divided responsibilities of cloud service providers (CSPs) and their customers. This means that each has their own tasks in maintaining security.

In short, the provider is responsible for securing the underlying cloud infrastructure, but the customer needs to secure their own data, configurations, and any applications they may have.

In more detail, the provider takes care of:

• Physical security: Secures physical data centers and hardware.
• Host infrastructure: Secures and maintains networking, servers, and storage.
• Data protection: Provides encryption for data in transit and at rest, but does not secure customers’ data.
• Applications: Potentially provides app hosting platforms or tools, but does not secure customers’ apps.

On the other hand, the customer secures everything they own:

• Data protection: Secures, classifies, backs up, and encrypts its own stored information properly.
• Information access: Manages secure access to sensitive data, user accounts, permissions, and multi-factor authentication.
• Configuration: Sets up firewalls and network settings properly.
• Applications: Develops, maintains, and configures apps and their secure application codes.

(Image credit: Future)The first step is defining it - now meet your provider

It would make things a lot easier if these rules and divisions were standardized, so when you understand them for one provider, you understand them for all. Alas, that’s not the case.

The above guidelines are general because the model somewhat differs between providers, and this lack of uniformity contributes to security gaps between the provider and the customer.

That said, make sure you understand your full responsibilities with each provider you use. For example, AWS says “customers should carefully consider the services they choose as their responsibilities vary depending on the services used, the integration of those services into their IT environment, and applicable laws and regulations.”

AWS makes a clear distinction between these responsibilities, saying that the customer is in full charge of managing the guest operating system, including updates and security patches. They must also configure the AWS-provided security group firewall and manage any other application software they own.

(Image credit: aws.amazon.com)

Microsoft further explains that responsibilities vary depending on where the workload is hosted: on software as a service (SaaS), platform as a service (PaaS), infrastructure as a service (IaaS), or in an on-premises datacenter.

However, “for all cloud deployment types, you own your data and identities,” it says. “You're responsible for protecting the security of your data and identities, on-premises resources, and the cloud components you control.”

(Image credit: learn.microsoft.com)Google Cloud’s ‘Shared Faith’

Google Cloud went with yet a third approach, differentiating between the shared model of other providers and its own “shared faith”:

Given that understanding the shared responsibility model properly in order to secure one’s data can be challenging, “Google believes that the shared responsibility model stops short of helping cloud customers achieve better security outcomes,” it writes. “Instead of shared responsibility, we believe in shared fate.”

The explanation of this approach focuses on “partnership” between the provider and the customer for increased security, as well as offering support and resources to help organizations secure data on their end.

The provider states that “a key component of shared fate is the resources that we provide to help you get started, in a secure configuration in Google Cloud. Starting with a secure configuration helps reduce the issue of misconfigurations, which is the root cause of most security breaches.”

(Image credit: Future)Why does shared responsibility matter, and what happens when companies disregard it?

The shared responsibility model is meant to provide a division of security, compliance, and ethical duties. The number one goal is to cover all bases to ensure all security controls are managed, thus preventing security gaps and potential breaches.

Should a breach happen, this shared control system is theoretically designed to help contain it faster and more efficiently. I say “theoretically” because security gaps do exist, as we’ll see later.

Additionally, the model is designed to ensure that neither the provider nor the client duplicates security steps unnecessarily, wasting their time and effort.

Finally, the security and infrastructure cost is much lower than managing one’s own cloud, and the deployment speed is much higher.

Overall, the model provides the benefits of the public cloud without having to maintain its infrastructure.

According to AWS, this shared model “can help relieve the customer’s operational burden as AWS operates, manages and controls the components from the host operating system and virtualization layer down to the physical security of the facilities in which the service operates.”

However, failing to follow it for any reason carries significant risks:

• data breaches
• cyberattacks
• financial loss (ransomware, regulatory fines, remediation costs)
• legal liability and potential lawsuits
• reputational damage.

The first is the one you should be the most worried about: should it happen, all the others will follow.

Simple misconfiguration mistakes can lead to data breaches, and unfortunately, this happens all the time.

Cracks in the model

This is a good time to discuss the issues with this system, and the first one is certainly the problem of misunderstanding, which leads to overlooked responsibilities and, subsequently, to security gaps in the customer’s databases, hence between the provider and the customer as well.

This issue results in misconfiguration - the direct culprit for most cloud security breaches.

Understanding responsibilities in a constantly changing environment is difficult. Sometimes organizations try to follow it, but overlook something. Sometimes they’re not even aware that they have this responsibility, presuming it all falls on the provider. And often they fail to protect certain data that may seem less relevant, such as backup files.

Either way, the end result is the same: databases are left unprotected and exposed for all to see and download.

This is so far from being theoretical that it’s nearly the norm. Just recently, security researcher Jeremiah Fowler discovered yet another unprotected database, with 666,369 records, including names, email addresses, phone numbers, IP addresses, and hashed passwords of the Tribeca Festival employees and attendees.

Potential consequences are massive and far-reaching for both the organizations that failed to protect the data and the individuals whose data was exposed.

Other issues in the shared model system include getting full visibility into the cloud environment, a lack of effective integrations that sustain and improve collaboration, challenges integrating tools and platforms, etc.

And if you’re managing security in a multi-cloud environment, that’s a larger beast to deal with.

Final thoughts

Protecting data is neither only the cloud service provider’s job nor that of the company. Rather, the job is divided between “mine” and “yours”. Generally speaking, providers protect what they’ve created, their house, so to speak. Customers are in charge of making sure all their own data and related services in the rented rooms are properly secured.

This is more easily said than done, given that, despite the many advantages of sharing this responsibility (including significant time and cost savings), mistakes happen a lot for many reasons, and the customers’ databases end up exposed. It’s a key issue to overcome if this model is to function successfully long-term.

In 2025, Hisense introduced RGB mini-LED to the market and since then, most TV brands have thrown their hat into the RGB TV ring. TCL, Samsung, LG and Sony are major names who have adopted the new technology, which was dubbed as being a potential ‘OLED killer’, thanks to boasting higher brightness and bolder colors, and claiming to deliver black tones to rival the best OLED TVs.

One of those new TVs has just entered our lab: the Sony Bravia 7 II. The more affordable model of Sony’s RGB TVs, sitting below the flagship Sony Bravia 9 II, the Bravia 7 II is priced fairly similarly to rival mid-range OLEDs. So, I decided to see how the Bravia 7 II would fare against the five-star LG C6, one of the best TVs I’ve tested in 2026 so far.

For these tests, both TVs were set to their respective, out-of-the-box Filmmaker Modes. For the Bravia 7 II, this was called Professional in SDR/HDR, and Dolby Vision Dark when displaying Dolby Vision content. On the LG C6, it's just the Filmmaker Mode.

Colors

While the Bravia 7 II (left) does deliver impressive colors with solid brightness, the C6 (right) colors look bolder, shown with the Lumas from The Super Mario Galaxy Movie. (Image credit: Nintendo / Universal Pictures / Future )

Starting with The Super Mario Galaxy Movie, both the Bravia 7 II and C6 demonstrated excellent color reproduction. In the opening scenes, the Lumas’ colors are richer and deeper on the C6, whereas they were brighter on the Bravia 7 II, which worked well for highlights on screen such as when Rosalina uses her wand.

While most colors were a bit punchier on the C6, I did find that yellows and golds actually appeared more vibrant and eye-catching on the Bravia 7 II, shown by the light from the fire reflecting in Rosalina’s hair and face.

Later scenes showed that the C6 had much bolder reds, shown in Mario’s hat and overalls. The C6’s colors also looked more accurate for the most part to me, though during an early scene where Mario and Luigi ride through the desert, I was drawn to the Bravia 7 II’s warmer color tone, with the desert sand looking more true-to-life on the Bravia 7 II, whereas it looked too red on the C6.

There was no denying though the C6’s color handling allowed for more texture and saturation within them, which worked well for this animation.

There's a nice balance to the Bravia 7 II's (left) colors in Speed Racer, but the C6 (right) has the vibrancy edge once again. (Image credit: Warner Bros / Future )

Moving to a movie with some real-life elements, Speed Racer, both TVs did a great job reproducing the movie’s over-saturated, cartoonish color profile. The ’60s-inspired, often-garish color of people’s outfits and the decor of the Racer family home really popped on both screens. S

imilar to Mario Galaxy, the C6 demonstrated deeper, bolder colors, while the Bravia 7 II showed brighter colors. In the scene in the park where Speed first meets Trixie when they’re both children, the red of Trixie’s dress is punchier on the C6, but the green grass is more vivid on the Bravia 7 II.

In some instances, the Bravia 7 II’s colors did look a little washed out, bordering on too-pale, which was exaggerated by comparing it directly to the C6’s color depth. Reds appeared slightly orange on the Bravia 7 II compared to the C6, a similar story to what happened when I compared the Samsung S95/S99H with the G6, with the former showing orange-tinted reds.

Still, color reproduction is one of the selling points for RGB TVs, and while I felt that the Hisense UR8 (that company's own more affordable RGB mini-LED TV) didn’t stack up that well for color against the Samsung QN80H standard mini-LED when I tested the UR8 and QN80H side-by-side, the Bravia 7 II is putting up a solid fight against the LG C6 OLED (which is a superior TV to the QN80H).

Contrast/dark room viewing

The C6 demonstrates deeper black levels in The Batman, with strong contrast, while the Bravia 7 II has a nice balance itself. A note: the C6 does not appear to have this much of a blue/green tint in person. (Image credit: Warner Bros / Future )

The Batman is something of a torture test for TVs due to its low brightness, which means dark tones can look inaccurate. While OLEDs do a better job displaying it generally, recent models such as the LG G6 have over-brightened the dark tones at times — but the C6 has always handled it well, so it's a great test against the Bravia 7 II.

Watching scenes from the Batcave early in the movie, both TVs demonstrated strong contrast with a great balance between light and dark tones. The C6 had a higher perceived contrast as its black tones were much deeper, thanks to OLED’s pixel-dimming abilities, which meant the dark surroundings that contrasted with the overhead lights and lights from Bruce’s equipment create a higher dynamic range.

Still, the Bravia 7 II demonstrated strong backlight control, better than almost any standard mini-LED TVs I’ve tested this movie on. I also praised the Hisense UR8 for this, so it certainly looks like RGB TVs have unlocked better contrast and dark area control for LED TVs.

Both TVs have strong contrast, and while the Bravia 7 II (left) has better balance in places, its backlight created some clouding that the C6 (right) did not suffer from. (Image credit: Warner Bros / Future )

In the subway fight scene, the C6 once again proved to have the better black tones, accurately capturing the gloomy tone of the film. However, on the Bravia 7 II, more details such as Batman’s armor and the texture of the back wall were legible, thanks to showing them in slightly higher brightness (more on that in a while), compared to the C6.

The C6 itself showed the occasional sign of black crush compared to the Bravia 7 II. For out and out contrast though, I preferred the C6’s look, but the Bravia II was impressive.

Watching in a dark room, the C6 has the benefit of no backlight, meaning there was no ‘cloud’ effect, where a backlight is obvious over a dark scene. While the Bravia 7 II did show the odd moment of clouding, again it showed excellent backlight control compared to the competition.

There was also very little blooming in the scenes I tried; it only became obvious when a bright menu appeared on the bottom of the screen. Even the Bravia 7 II’s viewing angles were solid, not showing anywhere near the level of blooming a standard mini-LED TV would when watched way off-angle.

Both TVs delivered accurate, rich black tones in this scene from Alien: Romulus but the C6 (left) has the advantage thanks to its self-dimming pixels. (Image credit: 20th Century Studios / Future )

I decided to try one final black level test, using the scenes of space from Alien: Romulus. As expected, the C6’s black levels were inky and deep, delivering strong contrast against the stars.

Again though, the Bravia 7 II’s black levels were very good for a backlit TV, although there was some graying on occasion.

I did notice more blooming on the Bravia 7 II in this film however, most prominently whenever bright, white text appeared against black backgrounds. Still, the Bravia 7 II fared better with this movie than I thought.

Brightness

The Bravia 7 II's much higher fullscreen brightness means large areas of the same color (the snow here) look much brighter than on the LG C6. (Image credit: Future)

I'm about to get into brightness measurements, but there's a caveat. For my testing, I profile a Klein K-10A colorimeter and Jeti Spectral 15VA spectroradiometer for the best accuracy — and for some reason, I could not get the two’s Delta-E results to line up completely when preparing to measure the Bravia 7 II.

The same equipment worked fine with other TVs when I checked it, so it suggests there’s something odd with the 55-inch Bravia 7 II unit I tested, or perhaps with the way the equipment interacts with it.

Before my full review, we’re getting a different, 65-inch unit in, and I will measure it again — however, I'll provide the measurements I took so far for reference, because while the Delta-E wasn't perfect, it wasn't too far off, so they should be instructive. (And the LG C6 measurements have no issues.)

Brightness results for these two TVs are actually quite surprising. In Filmmaker Mode, the C6 registered 1,427 nits of peak HDR brightness in a 10% window, while the Bravia 7 II registered 1,218 nits peak HDR brightness in its equivalent Professional mode. I was surprised to see that the C6’s peak brightness was higher, as I thought an RGB mini-LED would hit much higher numbers than an OLED.

Looking at fullscreen HDR brightness (a 100% window), this is where the real difference lies. The C6 hit 247 nits in Filmmaker Mode, while the Bravia II registered 957 nits in Professional mode. This is a huge difference — this is the area where mini-LED in general trump OLEDs, though 957 nits is especially high, even

Seeing how this translates into real-world viewing, it shows mostly in bright scenes or scenes with a lot of white. Watching snow scenes from the Spears & Munsil demo footage, a shot of some horses in heavy snowfall is significantly brighter on the Bravia 7 II compared to the C6, giving it a more realistic look.

Again, the Bravia 7 II benefits from higher fullscreen brightness for this image, even though the C6 can achieve higher peak highlights in places. (Image credit: Future)

Another shot of a cliffside during the daytime again looks much brighter on the Bravia 7 II compared to the C6. As the cliff takes up a large portion of the image, the Bravia 7 II, which is brighter in measured areas like 50%, 75% and 100% compared to the C6, means the image looks brighter overall on the Bravia 7 II.

In scenes from Speed Racer, the C6’s peak brightness means white highlights look glossier, such as on the shell of the Mach 5, and colors have more punch in highlight areas, such as the red M on Speed’s helmet. But as the Bravia 7 II has higher fullscreen brightness, the picture does look brighter overall on the Bravia 7 II.

Although the Bravia 7 II has higher fullscreen brightness, in this scene from Lawrence of Arabia, the brightness appears much closer between the two. (Image credit: Sony Pictures Entertainment / Future )

Finally, using a scene in the white-sanded desert from Lawrence of Arabia, this is one movie where the C6 and Bravia 7 II are much closer in brightness in real-world movies. As Lawrence rests on a rock, the sands behind him look punchier on the C6. His metallic watch had slightly more pop too, although this was minimal.

When the sands take up more of the screen, as a stranger approaches Lawrence and his guide, the C6 once again had higher perceived brightness for me.

In this instance, it seems to be related to both TVs color profiles. The Bravia 7 II’s warmer color tone, giving it the brown/orange tint you expect from this type of picture mode, blends more with the desert, whereas the C6’s red tone gives the desert a little more pop.

In terms of bright-room viewing, where you're contending with reflections, the Bravia 7 II is better overall, although the C6 does handle brighter scenes well, such as the ones from Lawrence of Arabia.

With darker movies, the Bravia 7 II is the better option, but it is still prone to mirror-like reflections (the Bravia 7 II doesn’t have the same anti-glare screen as the step-up Bravia 9 II). However, they weren’t as noticeable as the mirror-like reflections from the C6’s even glossier screen.

A step forward, but not quite there yet

The C6 (right) is still my pick, but this RGB vs OLED battle is getting more closely fought every time (Image credit: Nintendo / Universal Pictures / Future )

Whenever I’ve compared standard mini-LED and OLED, OLED has often come out on top, thanks to its richer black tones and contrast, more refined image and deeper, bolder colors. RGB has been touted as the latest big rival to OLED, and the Bravia 7 II gives the C6 OLED a really good fight.

When smaller RGB TVs first joined the market this year, with the tech itself only arriving in 2025, I thought prices would be inflated. While the C6 is the better priced TV right now, with a 55-inch model costing $1,499 / £1,499 / AU$2,995, a Bravia 7 II isn’t priced too differently, with a 55-inch model costing $1,799 / £1,799 / AU$2,995. (Though LG TVs tend to get more aggressive discounts than Sony TVs, so LG might retain the edge over time.)

For now, the C6 is still the TV I recommend between these two, because it offers the crisper look, bolder colors and more dynamic contrast I’m looking for, as well as a cheaper price. But, there’s a lot to like about the Bravia 7 II and if this is a sign of what’s in store for RGB TVs as the tech develops, OLED will need to watch its back.

Thinking of buying a new TV?

Try our TV size and model finder! You tell it how far you sit from your TV, we'll tell you what size to buy based on viewing angle advice from image quality experts, and we'll recommend our three top TVs at that size for different prices.

Square Enix is gearing up to kick off a brand new era of Final Fantasy 14 with the release of the Evercold expansion. The beginning of the "Godless Realms Saga" story arc, it arrives after the epic conclusion of a decade-long saga, and follows on from Dawntrail which laid the groundwork for this new adventure.

Game producer and director Naoki "Yoshi-P" Yoshida has already teased the new threat players will face in the DLC, which I'm beyond excited for, as well as a number of incredible quality-of-life system updates, new content, and improvements to outdated gameplay features.

It's looking to be a mighty expansion that will hopefully draw in even more players when the massively multiplayer online game (MMO) officially hits Nintendo Switch 2 next month, but there are a few things I'm particularly eager to experience.

Buckle up, we're saving the world again

(Image credit: Square Enix)

Final Fantasy 14 story spoilers ahead

After a decade of buildup, the Shadowbringers and Endwalker expansions saw the pieces of an overarching story finally come together and conclude the Hydaelyn-Zodiark saga, resulting in one of the best Final Fantasy stories ever to be told.

While some events were wrapped up with resolute endings, some lingering questions were left open that presented new possibilities for the MMO, and it all starts with the Reflections: mirror worlds created by the Sundering, a great calamity that saw the Source shattered into 13 Shards.

Players have visited other Reflections in past expansions, but in Evercold we'll visit the Fourth Reflection, which was beset by a calamity of ice. This means an entire new world to explore, and from the trailers and concept art, it appears that SE is leaning heavily into the fantastical style I adored in Shadowbringers and Endwalker, and drawing from what looks like Norse mythology for inspiration.

A new, unexpected threat to all Reflections has also emerged with the recent release of Patch 7.5, the Solstice, which will bring about the natural Rejoining of the Shards and threatens to wipe out all life on the Source and its Reflections.

(Image credit: Square Enix)

"In the time of the Ancients, Etheirys was divided across ten and three reflections to avert a terrible calamity," the expansion description reads. "Thus, for twelve thousand years did the star survive, and by its enduring hope did the Warrior of Light at last thwart its Final Days. Yet victory comes at a cost, and fate now demands its due.

"Faced with the new threat of the Solstice, the Warrior of Light is summoned to the Fourth reflection. But what do they hope to gain from wandering a world lost to ice?"

It's a bold move on the writers' part to thrust players back into a situation where they'll be dealing with the end of the world again; however, Dawntrail's subsequent patches have done a great job of building the foundations to make it a story worth telling.

I'm a FF14 player who looks forward to new story content most of all, and in Evercold, with the way the main scenario has been playing out, I'm certain our trip to the Fourth will open new avenues to explore and deepen the lore of the Ancients, which had been drip-fed to us over the years. In particular, the mystery behind the creation of Ethos, the key that opens up the path to these Reflections, and how this new ability to world-hop will play into the larger story.

Alongside these huge stakes, we've got a new "big bad" to face, the Winterers and the Ascian Halmarut, who are working in the shadows to prevent the Solstice.

Halmarut is unique in that she is sympathetic to mankind's current state and will do anything to ensure it survives the Solstice; however, her plan of controlled Rejoinings comes at the cost of mass death and destruction, believing that it's better to sacrifice some lives so humanity has a chance. But as the Warrior of Light, we can't go along with that idea.

Interestingly, unlike other Ascians before her, Halamurut encouraged the player to find their own way to deal with the Solstice. The Ascian was only introduced in recent patches, but given how superbly the writers developed its past villains (namely the best Final Fantasy baddie ever, Emet-Selch), I'm eager to see how SE will handle another multidimensional antagonist with a world-ending goal.

Raid content galore

(Image credit: Square Enix)

FF14 has always excelled at crossover raid content, whether that be the Yorha: Dark Apocalypse Alliance raid series, developed in collaboration with Nier developers Yoko Taro and Yosuke Saito, or the major raid series inspired by previous Final Fantasy games, like Final Fantasy 3, Final Fantasy 8, Tactics, and more.

Dawntrail's major Alliance raid series, Echoes of Vana'diel, wrapped up with the latest patch, a 24-player crossover celebrating Final Fantasy 6, that offered incredible bosses and a challenging yet rewarding experience.

For Evercold, SE revealed that the Alliance raid series will be based on the hit series Evangelion, titled "Ghosts of Desire", which acclaimed director and animator Mahiro Maeda will join as a special guest artist and designer. Like the Nier raid series, I'm expecting a high-quality experience with some unbelievable boss encounters that push the boundaries of the MMO; however, I'm more excited about the eight-player raid series SE has just announced at FanFest EU.

In the next expansion, FF7 Remake series director Naoki Hamaguchi and Yoshi-P took to the stage to unveil "Beyond the Lifestream," a FF7-themed raid series.

It feels like I've waited years for a proper FF7 collaboration, and the series will come just in time for Final Fantasy 7 Revelation, the final installment in the remake series.

Ahead of Evercold's release, FF14 will also be getting a FF7-themed adjacent feature in the MMO's Gold Saucer on October 28, Keybound Brawler, a new mini-game that will reward players with a glamour bundle inspired by Tifa and Barret's outfits from FF7; a teaser for what's to come.

While SE has yet to share the full raid details, I'm expecting the usual crossover gear (I'm begging for Cloud, Sephiroth, and Aerith outfits, please!), elaborate boss fights inspired by some of FF7's most iconic moments; a confrontation with Safer-Sephiroth while a remixed medley of the FF14 and One-Winged Angel themes plays in the background is surely a given.

Raid maps are also always incredible in FF14; I remember my first time entering the first raid of the YoRHa series and being blown away by the scale of the set pieces and attention to detail, so with Hamaguchi's team working alongside the FF14 team, I'm anticipating something next-level; hopefully a recreation of the dystopian metropolis of Midgar?

It's time for improvements

(Image credit: Square Enix)

FF14 has been running for more than a decade after it was rebooted into A Realm Reborn in 2013. Since then, the MMO has seen innumerable changes, including to Jobs, a huge graphics overhaul that modernized the game's visuals, not to mention the upcoming expanded character customization options that players have been yearning for.

But some things have always remained the same, primarily the battle system and activity features. With the release of Evercold, FF14 will receive a significant overhaul with the introduction of the 'Reborn' and 'Evolved' modes.

Reborn mode will be based on the current combat system and Job mechanics, while Evolved mode offers greater emphasis on "job identity", updating combat rotations and changing some abilities, and will only apply to existing Jobs.

It's an exciting new change, because while Jobs occasionally get rebalanced, this is the first time the game is getting a significant gameplay revamp that will allow long-time players to experiment with a new way of playing, and newcomers have two options at their disposal.

What's more, the expansion will also see an overhauled gear progression system with the introduction of 'Adventurer Activity.' In its current state, FF14 requires players to log in every day to complete daily roulettes to earn XP and accrue their weekly cap of 450 Allagan Tomestones, which are exchanged for high-level gear; however, this can be a downside, especially for players who can't log in every day.

(Image credit: Square Enix)

With Adventurer Activity, this redesigned framework will change from a daily to a weekly-based reward system, remove the exhausting grind of completing daily roulettes, and instead offer more player freedom.

Adventurer Activity (which may look like a battle pass system, but Yoshi P insisted it absolutely is not!) will let players complete a series of missions to amass activity points instead, whether it be dungeons, PvP, Trials, and more, and SE is even considering allowing unearned points from previous weeks to be collected if players miss out.

This is perfect for people like me who sometimes don't really feel like running roulettes and also gives players a chance to explore other content they wouldn't normally engage with.

Final Fantasy 14: Evercold will arrive in January 2027 for PC, PS4, PS5, Xbox Series X and Series S, and Nintendo Switch 2.

The “Seems like AI slop” button will help LinkedIn train its AI-detection tools.

• Microsoft launches MAI-Cyber-1-Flash, its first in-house cybersecurity model.
• It also reveals Project Perception, an agentic system whose red, blue, and green agents find, triage, and patch vulnerabilities
• The launch comes days after OpenAI said its models escaped a sandbox and attacked Hugging Face

Microsoft has unveiled two significant security announcements - MAI-Cyber-1-Flash, the first cybersecurity model it has trained in-house, and Project Perception, an agentic defense system that finds vulnerabilities, decides which ones matter, and writes and deploys the patches.

The launch took place days after OpenAI disclosed that its own models had broken out of a sandbox and hacked Hugging Face, which makes the timing either unfortunate or pointed.

With Microsoft claiming a 96% score on CyberGym, an industrial benchmark for cybersecurity, nearly 12 points above Anthropic's Claude Mythos 5, while promising as much as 50% savings in token cost, the company says it is bringing what it calls a 'well-tuned, multi-model system with access to uniquely rich historical training data',

A competitive product with excellent timing?

The Hugging Face incident drew much attention, not all of it was negative: as it painted OpenAI's models as more capable than the company's own assessments had suggested. It also sharpened a question researchers have been raising for years: what happens when a model becomes capable enough to defeat the controls placed around it?

Microsoft's answer is well-timed and two-pronged, claiming to be both cheaper and more capable than the alternatives, though the performance figures are, so far, Microsoft's own.

MAI-Cyber-1-Flash, is a sparse mixture-of-experts transformer with 137 billion total parameters, five billion active, and a 256,000-token context window, built as a cybersecurity fine-tune of MAI-Code-1-Flash, itself developed from a MAI-Thinking-1 mid-training checkpoint.

It is designed to handle up to 90% of the tasks inside MDASH, Microsoft's multi-model vulnerability harness, with OpenAI's GPT-5.4 reserved for the hardest 10 percent.

Microsoft says that split costs about half as much as its previous best MDASH configuration. For now, it runs only within MDASH and is available to approved MDASH customers through an Azure AI Foundry private preview, with no standalone API.

Project Perception is the wrapper around it, drawing on MAI-Cyber-1-Flash for its first workflow alongside frontier models such as GPT-5.4. Three classes of agent split the work: red agents probe for paths an attacker could take, blue agents investigate and decide what constitutes meaningful risk, and green agents remediate and harden.

A chart from Microsoft showing how Project Perception splits security and operational tasks across coordinated multi-agent teams (Image credit: Microsoft)

On the specific failure that let OpenAI's models loose, Microsoft took the precaution OpenAI did not, as it says all benchmark testing ran in a network-isolated environment with no access to production systems, the public internet, or external services.

OpenAI's sandbox, by contrast, kept one route outward in the form of an internal package-fetching service, and its models found a flaw in it, escalated privileges and worked across the research network until they reached a machine with internet access. Microsoft says its isolation held. Nobody outside Microsoft has verified that.

The harder question is not containment during testing. Project Perception moves the work out of a research network and into customer production environments, where green agents are authorized to change live systems as their normal function.

There is no sandbox to escape, because acting on real infrastructure is the product. And attribution is difficult even when someone is watching: Hugging Face detected the intrusion within days and reported it to law enforcement, but had no idea who was behind it until OpenAI said so.

It also could not get help from the leading American models, which read its defensive requests as offensive ones and refused. It ended up defending itself with GLM 5.2, a Chinese open-weight model, running on its own infrastructure.

For now, the industry's answer to dangerous capability remains narrower distribution, and the one documented case of a defender needing that capability urgently ended with them reaching for a model nobody had gated.

While AI is shaking up all industries, few have been impacted as visibly or as quickly as web design.

No longer can agencies charge for time on repetitive tasks that can easily be farmed out to AI. This creates a challenge for businesses that need to prove their value to clients that are now eyeing up cheaper, faster solutions to their needs.

I caught up with Itai Sadan, CEO and Co-founder of dedicated agency website builder Duda. We talked about Duda's new vibe coding tool, how it fits into his vision of the future of website building, and how he is seeing agencies demonstrate value through consultation and creative talent in 2026.

Firstly, tell us a little bit about the new AI tools you have rolled out across Duda. Why has Duda focused on launching these tools for agencies now?

AI has lowered the barrier to creating a website, but it has not solved the challenges agencies face when delivering websites at scale.

Agencies still need to produce work that reflects a client's brand, collaborate efficiently across teams, maintain quality standards, and do it all within a business model that supports healthy margins. We built our AI capabilities with those realities in mind, rather than adapting generic AI products that were never designed for professional web creators.

Our approach combines the speed of AI generation with the capabilities agencies already rely on in Duda.

Website generation turns a prompt into a production-ready starting point while preserving access to Duda's drag-and-drop editor, performance, SEO foundations, and the flexibility to refine every detail before launch, either by prompt or by simply clicking with a mouse, which is many times faster and more cost-efficient.

Duda Vibe expands those possibilities with an AI-native environment for creating websites and applications that remains fully connected to the Duda platform.

Agencies can collaborate with their teams and clients, white label the experience under their own brand, and operate with the predictable pricing and workflows they depend on to scale profitably. Our goal is to help agencies take full advantage of AI while strengthening the systems and processes that make their businesses successful.

AI tools, like those you have just rolled out, remove considerable admin work within agencies; should agencies be adjusting pricing models to reflect this?

In many cases, agencies will have an opportunity to deliver greater value than before, and their pricing should reflect the outcomes they create, not the time they save.

Itai Sadan, CEO and Co-founder of Duda

I don’t think AI changes the fundamental value proposition of an agency. Clients aren't paying for time spent on repetitive tasks. They're paying for expertise that produces measurable business outcomes, whether that's more qualified leads, more appointments, or more online sales.

There's no shortage of website builders, and now even AI coding tools, built specifically for small businesses to use themselves.

What an agency brings is the ability to translate business goals into digital results. That includes understanding how a website should be structured, how it should be written to attract and convert visitors, and how it fits into a broader marketing strategy that may include SEO, paid advertising, social media, and other services. AI can accelerate many parts of that process, but the quality of the outcome still depends on the experience and judgment of the agency using it.

When I speak with agencies, they don't tell me they're working less. They tell me they're spending their time differently. The hours saved through automation are being reinvested into higher-value work, whether that's providing a more consultative client experience, expanding the services they offer, or helping clients adapt to new opportunities like AI search.

That's been Duda's philosophy from the beginning. We've always focused on removing the repetitive work so agencies can spend more time on the work their clients actually value.

As AI continues to improve efficiency, I believe that principle becomes even more important. In many cases, agencies will have an opportunity to deliver greater value than before, and their pricing should reflect the outcomes they create, not the time they save.

AI can make errors. For example, it may interpret a brief incorrectly, which could lead to significant additional work for agencies to fix. How much oversight do you recommend agencies give AI tools? Are there systems and processes you recommend to limit potential issues down the line?

I often compare AI to an exceptionally talented intern. Give that intern a clear objective, the right context, and the supporting materials, and they'll produce impressive work.

Itai Sadan, CEO and Co-founder of Duda

We've tried to address that challenge at the platform level.

With Duda Vibe, for example, the planning stage interviews the agency before it begins generating anything. That additional context consistently produces better results because the AI is working from a clearer understanding of the project's goals and requirements.

We're also continuously improving the models behind our AI website generation so they can interpret briefs more accurately and produce websites that require less refinement before launch.

That said, AI should never be treated as a substitute for professional review. We recommend that agencies establish a quality assurance process for every project, just as they would with work created entirely by humans.

Duda's AI Copilot can assist with that review by identifying areas for improvement before a site goes live, while our Optimization Center continuously audits websites for common SEO, accessibility, and AI search optimization opportunities that might otherwise be overlooked.

AI is most effective when it's used throughout the entire workflow, not just at the point of creation.

I often compare AI to an exceptionally talented intern. Give that intern a clear objective, the right context, and the supporting materials, and they'll produce impressive work. Give them a vague brief, and they'll fill in the gaps with their own assumptions. That's why we encourage agencies to be as specific as possible.

Duda supports rich prompting with reference documents, images, and other project assets so agencies can provide the context the AI needs from the very beginning. The better the guidance, the better the outcome.

Vibe coding tools like Bolt of Lovable make it easy for anyone to build web apps, but lack agency-level infrastructure. How is Duda Vibe tailored to better meet agency needs compared to generic vibe coding tools?

Most vibe coding tools are designed to help an individual create an application as quickly as possible.

Agencies need much more than that. They need a platform that supports collaboration, client management, security, governance, and long-term maintenance across dozens, hundreds, or even thousands of projects. That's the gap we set out to address with Duda Vibe.

Duda Vibe is fully integrated into the Duda platform, giving agencies a centralized workspace where they can manage every project, whether it was built with our traditional website builder or in Duda Vibe.

Teams can collaborate throughout the development process, clients can securely review their own projects through a fully white-labeled experience, and agencies can continue using the workflows they're already familiar with.

Beyond development, Duda Vibe includes the operational capabilities agencies depend on, including automatic version history, secure data storage, reliable hosting, and performance insights that help them monitor both site performance and AI visibility over time.

We're still in beta, but our roadmap is firmly focused on the needs of agencies and the way they work.

One feature I'm particularly excited about is bringing our popular inline commenting workflow to Duda Vibe. Clients will be able to leave feedback directly on the project, just as they do today in our website builder. Once the agency reviews and approves a request, Duda's AI Stack will understand the context of that comment and carry out the requested changes automatically. It's another example of how we're applying AI to reduce the operational overhead of working with clients, not just the time it takes to build an application in the first place.

Although great for quick builds, vibe coding apps can lack the architecture, stability, and security required for long-term success. How does Duda Vibe help agencies maintain web apps without creating technical debt?

The technical debt challenge with AI-generated applications usually isn't the code itself. It's everything required to run and maintain that application over time.

Itai Sadan, CEO and Co-founder of Duda

The technical debt challenge with AI-generated applications usually isn't the code itself. It's everything required to run and maintain that application over time.

Many AI coding tools can generate a working prototype in minutes, but agencies are often left responsible for hosting, security, authentication, payments, infrastructure, and long-term maintenance. That's where technical debt begins to accumulate.

Duda Vibe takes a different approach. Every application is built on Duda's managed platform, so agencies have a foundation they can depend on long after the application is launched. Hosting, security, DDoS protection, high performance on metrics such as Core Web Vitals, platform updates, and other operational capabilities are built in from the start.

We also provide native services for capabilities like authentication, eCommerce, payments, and other backend functionality, allowing agencies to rely on platform services rather than AI-generated implementations for some of the most critical parts of an application. That reduces complexity, improves reliability, and significantly limits the technical debt that can accumulate over time.

Our view is that AI should accelerate development without shifting long-term maintenance risk onto agencies. The platform should handle the operational complexity, allowing agencies to focus on building applications that deliver value for their clients.

The ability to create custom widgets opens the door to creativity. What tips would you give agencies that want to use this feature to create ‘signature’ widgets to differentiate themselves from competitors?

Clients generally don’t know what their agency’s website builder can and can’t do out of the box, so it’s important for agencies to not get too caught up in individual widgets.

What makes a great website is how all of those widgets come together on the canvas and how well that collage of widgets meets the needs of the client. A great way to differentiate is to focus on the latter: the client’s needs.

For example, one agency building on Duda used the Widget Builder to generate an interactive carousel of 3D beer cans for a brewery. Not only did this look great, but it really elevated the overall design of the website. It was a nice touch that showed a level of care that I’m sure the client appreciated.

For agencies looking to stand out, I’d recommend double-clicking into the needs of the client. Breweries are in the hospitality industry, so a level of showmanship is beneficial online, but that same widget showing a rotating carousel of 3D toothbrushes would be odd to see on a dentist’s website. Understanding your client’s industry, then building widgets specifically for that vertical, is a great way to stand out and showcase expertise.

What do you think the biggest challenge will be for agencies when it comes to using AI over the next 12 months? How do you recommend agencies overcome these challenges? 

Agencies need to be able to produce something better than if an owner asked Claude to create a website for their business—which is not as easy as it sounds.

Itai Sadan, CEO and Co-founder of Duda

Agencies are being impacted by AI from dozens of different angles. Whether that’s changes to the way they work or the rise of AI search impacting how their clients get found, it’s an extremely adaptive and exciting time to be an agency. Staying ahead of it all is going to be the biggest challenge.

Take website building, for instance. The floor is going to continue to rise. Agencies need to be able to produce something better than if an owner asked Claude to create a website for their business—which is not as easy as it sounds.

Platforms like Claude are getting better each and every day. Whether that’s a better experience, a better product, or both, agencies need to be prepared to add something special that AI can’t offer by itself.

The hot topic — no pun intended — in haircare gadgets right now is heat. More specifically, the evils of heat, and how it's a one-way street to fried locks and frizz. So when I spoke to Dr Julie Chung, co-founder of haircare brand T3, I was surprised to discover she has a rather different take.

"Heat in general has been sort of demonized incorrectly," she tells me. "Women always think, 'Oh heat is terrible for my hair, I'm just gonna let it air dry, I'm gonna shower and just sleep on it.' Hair is actually in its most fragile state when it's wet. It tends to be much easier to break — that tensile strength is decreased and it's more likely to snap. And not to mention when you air-dry your hair, usually the hair is very frizzy and it tends to look even unhealthier if your hair is already in a dried out state."

Hair is in its most fragile state when it's wet. It tends to be much easier to break — that tensile strength is decreased and it's more likely to snap

That focus on hair health is nothing particularly new. "Women have always been obsessed with trying to find the healthiest option," Dr Chung tells me. "You know, people pursuing this idea of healthy heat is something I'm seeing [right now], but that's something that I've seen for 20 years."

However, the conversations have been buoyed up by the arrival of air stylers like the Dyson Airwrap. They've been heralded as the solution to all our heat-damage fears, but Julie is skeptical about that, too.

"The air styler category has grown quite big and there's this notion that air curling is better than curling iron curling, which I don't necessarily agree with," she continues. "I think it's right for some women in terms of ease of use, but I think curling irons can be just as healthy as air styling. It really depends on how many passes you go through in the hair, how adept the woman is with her styling."

T3's Aire 360 has a proprietary ceramic coating designed to deliver even heat (Image credit: Future)

So hot tools aren't inherently terrible, which is good news, because — as Dr Chung is at pains to point out — air stylers won't work for everyone. If your hair type isn't the kind to happily hold a curl that's been created by just a waft of a breeze, and you do decide to swap your air curler for a hot tong or enlist the help of a straightener to create ribbon curls, then it's worth finding a good one. Julie suggests paying attention to the kind of heat and how it's being delivered.

"The key there is ceramic. So in our Aire 360, for example, we have a ceramic coating instead of plastic because we know that ceramic is a better conductor of heat," Dr Chung explains. "We have a specific proprietary ceramic blend that evenly heats the plates without any hot spots. So it has a gentle delivery of heat along with far-infrared heat, which is a gentler heat as well."

Big air, better hair

What about hair dryers — those are safe, right? Wrong. Here, too, Dr Chung has opinions that seem to butt up against those of other makers, with their fixation on packing the most power possible into their dryers.

The average woman cannot and should not be using high velocity heat every day

"The biggest secret sauce — which isn't even a secret because I talk about it so openly, but even now our competitors do not want to embrace — is the fact that you do not need heat to be at the highest setting, and you do not need the highest velocity," she insists. "This is something that stylists sometimes need in order to do a quick blowout very quickly, but the average woman cannot and should not be using high velocity heat every day."

Dr Chung talks about how people will often add a concentrator "without really thinking", then use a round brush and really focus the heat in, with the aim of smoothing out the hair. "It's extremely damaging when done repeatedly", she warns.

Similarly, if you regularly pop your dryer on full-blast when rough-drying your hair, so it feels "like a wind tunnel where your hair is flying all over the place", that's not great either — and not just because the wind tunnel effect will kick up static.

(Image credit: Future)

"It'll dry your hair, yes, but will make it extremely dry and not just dry your hair, but take all the moisture out of your hair without sealing the cuticle," she explains.

The brand has found an innovative way to avoid this, without sacrificing efficiency, and the concept dates right back to T3's debut product. "We had this idea called Soft Air, where we delivered a larger volume of air at lower heat, so it would dry your hair very, very quickly, but without stripping it of all its moisture."

She explains how the team shaped the concentrator and the barrel of the dryer so that the air wasn't expelled in a dense stream, but instead came out in a larger volume, occupying a larger space. The idea of Soft Air is that it could still dry hair efficiently, but without the need to max out on power or temperature, and when combined with ions helps to decrease static frizz.

I've been trying out one of the brand's dryers (the T3 Afar) and the air delivery feels noticeably gentler — almost unsettlingly so — and less concentrated than any other hair dryer I've tested. It's slower to dry my hair than the likes of the Dyson Supersonic Nural, but it does leave my locks feeling softer and with less frizz than usual. So maybe Dr Chung is on to something. Excuse me while I dig my curling tongs out of storage.

In a world where data leaks are a constant, misconfigured databases remain the number one cause, exposing sensitive data to anyone who knows where to look.

This unfortunately common human error leaves the affected individuals and organizations open to hackers, cyber threats, scams, and further data leaks.

Even now, as you read this, there are plenty of companies that forgot to, didn’t know they had to, or didn’t care to lock their databases, leaving them open to the wide web.

Defining misconfigured databases

When we talk about misconfigured databases, we talk about systems that somebody set up without appropriate security settings, leaving them public for all the Internet to see (and potentially exploit).

Access controls may be missing or given to too many people, and default passwords/configurations could be left unchanged by the user.

Additional common issues include giving full administrative control to users and apps where basic is sufficient, storing sensitive information in plain text, and disabling security logs (which prevents tracking unauthorized access).

Everyday Jane and Joe, unfortunately, have no idea how common this issue is – and it’s their data on the line. We’re talking about tens of thousands of major security breaches in a matter of years, and there’s no estimating how many publicly accessible databases are still out there. Here are just three of these many examples.

In 2026, researchers found that European cloud provider Nextcloud kept an unprotected database on the public internet, containing 367,000 records (8GB) of sensitive employee and client data.

In 2025, IMDataCenter, a Florida-based data hygiene, enhancement, and append services provider, was leaking 38GB of sensitive personal records. The unencrypted and non-password-protected database held 10,820 in total.

In 2024, sports analytics technology company TrackMan exposed sensitive customer data: 110TB and 31,602,260 records. The database had no password.

The Tribeca Film Festival case

Notably, the above-mentioned examples were the whitehat work of cybersecurity researcher Jeremiah Fowler, who recently uncovered a new data leak affecting the Tribeca Film Festival.

Fowler found four publicly accessible databases without passwords or encryption. These contained 666,369 records in total, timestamped from 2019 to 2026. Among press kits and marketing materials were also full names, email addresses, phone numbers, IP addresses, and hashed passwords.

Some of these are huge names in the film world, including famous directors, producers, and actors.

A found spreadsheet listed data obtained through device/user fingerprinting, such as user-agent data, device information, browser details, and source applications.

Fowler couldn’t say who exactly managed the databases, how long they had been public, or if anybody else had accessed them before him. This is often the case.

(Image credit: Jeremiah Fowler, expressvpn.com)The issue of neglected backup files

Backup files often become forgotten, ignored, or simply deemed irrelevant, even when they contain a significant amount of potentially sensitive data concentrated in one place, aka a scammer’s treasure-trove.

Fowler describes these as some of “the most overlooked assets when it comes to data security.” While organizations often focus on protecting production systems, they put backups, exports, archives, or development environments on the back burner.

Yet, “backup files can become a single point of exposure if improperly secured,” the researcher warns.

The US case

One may think that it’s only private companies, perhaps smaller ones, that make these seemingly dumb human errors. And yet, the biggest breaches that happened as a result of misconfigured databases all belong to governments.

The United States is often cited as being at the top for these data incidents. Among other events, in early 2011, the Texas Comptroller’s Office revealed a breach of 3.5 million people’s personal information, including Social Security numbers, dates of birth, and driver’s license numbers. The office kept the information on a publicly accessible state server.

In late 2015, computer security researcher Chris Vickery found a database with 191 million voters’ information, 300 GB in total, available for all to see. The cause, you’ve guessed it, is an incorrectly configured database.

“If you know the IP address, boom, you can access all of it,” Vickery said. “No authentication, no password, no security whatsoever involved.”

(Image credit: Reddit)

In 2017, internet security firm UpGuard revealed that Vickery found a 1.1 TB-heavy database of the Republican National Committee (RNC) with personal details of nearly every registered voter in the U.S., collected on behalf of Donald Trump’s election campaign. It was publicly available for two weeks due to an improperly configured security setting.

The Brazil Case

Brazil's public sector has also seen a notable number of data exposures stemming from human error, not active hacking.

In 2018, cybersecurity researchers discovered a gigantic, openly accessible database with sensitive information of 120 million Brazilians, more than half of the population. A simple renaming mistake left the directory structure unprotected, which opened the database for viewing and downloading on the web.

2020 saw two significant leaks. Developers left database login credentials hardcoded and weakly encoded in the source code of the Ministry of Health’s official government website. The personal data and medical logs of more than 243 million people were left public for over six months.

Another leak that year exposed more than 16 million COVID-19 patients’ medical records after an employee uploaded a spreadsheet containing sensitive data on GitHub.

Potential consequences of misconfigured databases

The issues that could arise from misconfigured databases are manifold, with potential consequences hitting in waves years after the initial leak.

Businesses may see financial losses, regulatory fines, lawsuits, operational disruption, and reputational damage. Should hackers find these databases, they may demand payment to restore access, or they may sell the info on the black market.

For an individual whose data has been exposed, the consequences can be severe. Bad actors can abuse individuals’ personal data and financial details, and even use it to scam other people.

Let’s take the above-mentioned cases as examples. Publicly exposed internal data of a large festival could give criminals detailed info on the operations behind the scenes, as well as on the participants.

“This type of information could create privacy and security risks by revealing personal contact information that could be used for phishing, social engineering, identity correlation, impersonation attempts, or further reconnaissance of internal storage systems,” Fowler writes.

Many more people can be targeted with celebrity frauds, where scammers impersonate celebrities, using data as ammunition.

In the US voter data case, it’s not just a matter of each person’s individual data, but the sheer magnitude of the information concentrated into one database. This is very expensive and time-consuming for cybercriminals to do, but it’s incredibly useful for a wide range of fraudulent schemes.

For example, “if a group gets a hold of this, they have phone numbers for the rest of their lives to call,” Vickery said.

UpGuard argued that the fact “that such an enormous national database could be created and hosted online, missing even the simplest of protections against the data being publicly accessible, is troubling.”

“Beyond the almost limitless criminal applications of the exposed data for purposes of identity theft, fraud, and resale on the black market,” UpGuard writes. “The heft of the data and analytical power of the modeling could be applied to even more ambitious efforts – corporate marketing, spam, advanced political targeting.”

As for healthcare-related breaches, these databases can earn sellers a lot of money on the black market, given the amount and the sensitivity of personal information that can be exploited. The risks include identity theft, account takeovers, financial fraud, and creating additional profiles to scam others, indefinitely.

Cybercriminals can even blackmail patients and healthcare providers, typically asking for large sums of money.

Insidiously, the exploit could stay hidden, or the full scale of the consequences may not be visible right away, but weeks or even months later.

Shared responsibility model

Finally, despite the popular misconception, it’s not Amazon Web Services’ or Microsoft Azure’s duty to protect anyone’s databases. The provider secures the underlying cloud infrastructure, while the customer needs to protect their own data.

Based on the shared responsibility model, the customer must ensure the correct database configurations, securing all data properly.

Amazon writes that the customer assumes responsibility for any updates and security patches, as well as associated application software and the configuration of the AWS-provided security group firewall.

The chart below shows the Security “of” the Cloud versus Security “in” the Cloud.

(Image credit: aws.amazon.com)Final thoughts

Unsecured databases are the number one cause of data breaches and leaks, spanning industries and governments, which many believe ‘should know better.’ These databases usually contain a variety of sensitive information, inadvertently exposed to the open internet, without any basic security measures being taken.

This makes them a juicy target for hackers, who may utilize them for ransomware, sell them on the black market, or use them for a variety of scams.

Proper database configuration is each database owner’s job.

The year 2030 bears a lot of weight – not only does it mark the end of the current decade, but it marks a major turning point for sustainability across several industries and nearly all countries. Of most relevance for this discussion are the carbon, emission and water neutrality goals that global companies have set themselves.

By 2030, Google intends to reach net zero, Microsoft wants to become carbon negative and Apple wants to make its entire footprint carbon-neutral, but these are all goals that were set as far back as 2019 and 2020.

A lot has changed since then, and nobody could ever have imagined just how much generative AI was going to blow up when ChatGPT launched in late 2022, let alone the impact that agentic AI continues to have.

With 2030 previously envisioned as some sort of a finish line, that deadline is now under immense pressure as hyperscalers expand at an incomprehensible pace just to keep up with cloud and compute demands. So can climate commitments that were designed before the AI boom survive the infrastructure demands it's created?

AI has fundamentally changes sustainability progress

AI's impact on the chips market is already well-reported. Omdia's PC and tablet research director Ishan Dutt explained to me that an ongoing "capacity reallocation" is currently pushing memory makers' "DRAM/NAND wafer capacity toward HBM and high-capacity DDR5 for AI data centers," which is of course constricting supply for consumer devices.

Even without being familiar with the intricacies of this shift, consumers are already acutely aware of the ongoings due to sharp rises in PC, smartphone and storage prices.

What I'm more worried about is the long-term sustainability of AI-related expansions – while mounting local opposition is already calling hyperscalers and developers out over intense electricity and water consumption, this doesn't begin to cover the longer-term impacts. Upcoming 2030 deadlines are a great opportunity to benchmark current progress.

I'll explore four key areas of this in further detail with the help of Google's Director of Sustainability for EMEA, Adam Elman: grid capacity, water consumption, the efficiency paradox and local socioeconomic impacts.

The grid is the first bottleneck

In our exclusive interview, Elman acknowledged that recent "hyper-growth" has strained progress toward 2030 goals, but for Google in particular, the targets were intentionally "ambitious" to "push the frontiers of what is possible in energy systems and data center operations."

I questioned recent shifts in the trajectories of many companies' sustainability reports, including but not limited to Google's, such as spikes in energy consumption and emissions. Elman warned me that "the path to achieve these ambitions is not linear," implying that companies will constantly make adjustments as the landscape evolves.

Google saw a 37% rise in annual electricity demand, but still succeeded on 100% matching with renewable energy purchases to maintain momentum toward its goal. Amazon also saw a 16% rise in its last-year total emissions, and a 34% rise in purchased electricity emissions.

These figures are widely reported in publicly available sustainability reports, but I asked Elman whether companies should be more transparent about the finer details. He agreed that being open "about the complexity of the energy transition" amid this turbulent AI era is "the right thing to do," noting that this honesty can also help "drive the policy, market and technology developments needed to unlock an abundant clean energy future."

The company also stresses the importance of distinguishing between global annual offsets and true local grid-level hourly matching so that customers, investors and policymakers can understand the challenges.

(Image credit: May Cloud/Unsplash)Global warming is highlighting water pressures

I also asked whether water self-sufficiency is a realistic goal amid rising global temperatures, to which Google's response was overwhelmingly positive. Data center projects can follow one of many routes to reduce their impacts, including adopting closed-loop systems or using air cooling where that may be more suitable.

To relieve pressure on local water tables, Elman also told me that companies can opt to use recycled wastewater in closed-loop systems to get all the benefits of effective water cooling without consuming drinkable, fresh water.

Similar to energy reporting, Google supports local, site-level water reporting so that local citizens and utility companies can better understand the demands of individual campuses.

Efficiency drives total reductions

The reality is that emissions will inevitably rise as hyperscalers built out more and more data centers, but Elman told me that prioritizing absolute reductions or improving efficiency would be a false dichotomy: "True leadership requires prioritizing both absolute reductions and massive efficiency improvements."

Efficiency is an important part of the equation, and decoupling compute growth from resource consumption means companies can understand progress.

I asked about installing smaller, hyperlocalized data centers as a measure to tame local opposition, which I'll also explore below, but from an energy standpoint, Elman reminded me that larger facilities are generally more efficient because they can use large-scale cooling systems and support utility-scale energy agreements.

With ideas now starting to circulate around positioning data centers in the sea or as far as space just to improve thermal and power efficiency, Elman admitted that technical feasibility and practical implementation are two, very different things.

Google is now one of a growing list of companies looking at low-Earth orbit data centers, with Project Suncatcher observing an 8x energy per unit area increase compared with terrestrial solar panels. Prototype TPUs are set to arrive as soon as 2027, but as for wide-scale deployment, we're likely years or decades off the real deal – if at all.

Responding to local opposition

Public opposition is also on the rise amid wide-scale and local sustainability concerns, with many US states and cities already having imposed temporary memorandums to ban new projects as they get to grips with policies.

Google's EMEA Director of Sustainability told me that data center operators need to change how they've perceived, shifting from resource consumers into "supportive community anchors."

Some measures that Google suggested including paying for grid upgrades to protect local households from rising energy bills and supporting water projects to offset consumption. It's also not uncommon for companies to offer other local rewards to encourage citizens to accept their plans, like job opportunities, training schemes and other educational funding.

Is 2030 still a finish line, or the biggest test?

Ultimately, 2030 should not become an excuse for hyperscalers to withdraw or reframe the targets they set before the AI boom – instead, it should serve as an accelerant to encourage progress despite the challenges.

The technology has not made those earlier commitments obsolete, but AI has exposed how some earlier measures might not be so effective anymore.

Key to meeting these targets is openness and transparency around reporting for the purpose of policymaking, but it's also clear that one route won't solve the problem. Instead, success will come from endless components that will all add up to a meaningful result, be it driving efficiency, encouraging local support, responding to grid challenges, changing water habits or something else.

Whether or not the 2030 goal remains a finishing line is yet to be seen, but there's clear optimism that it's still a realistic goal for many.

AEO (Answer Engine Optimization) is the process of optimizing your online presence to get mentioned, cited, or recommended by AI platforms like ChatGPT, Gemini, or Perplexity.

It's no easy feat, but AEO tools can give you the insight you need to do this.

I've tested seven of the top AEO tools on the market, and HubSpot earns my top spot. This is thanks to its deep diagnostic tools, AI analytics, and CRM integration – it's also affordable, making it ideal for everyday use.

But an AEO tool isn't one thing. It's a label stretched across very different jobs.

Some tools only tell you whether AI mentions your brand. Some find the questions your customers are asking these platforms. Some show you which websites they trust and pull from. Others write and optimize the content that gets you picked in the first place.

I tested each tool the way you'd use them in real work. Same setup, the same kinds of prompts, the same questions put to each tool. I scored each based on tracking depth, setup time, insight clarity, ease of use, and value for money.

Below, I dive into my experience with these tools, giving you all the information you need to pick the best one for you.

Start your free 28-day trial and find out if your brand is the answer AI gives buyers

HubSpot AEO gives you visibility tracking, competitor analysis, citation analysis, and prioritized recommendations across ChatGPT, Perplexity, and Gemini — all in one place.

It's one of the fastest ways to understand where your brand stands in AI-generated answers and what to do about it.View Deal

Key takeaways

• If AI doesn't mention you, a competitor gets the customer: More buyers now ask AI before they choose.
• "AEO tool" isn't a single category: It spans four jobs: tracking visibility, finding prompts, showing trusted sources, and creating content.
• Optimizing for AI is still just SEO: There's no secret algorithm, and no tool is a shortcut past the fundamentals.
• Check what marketing pages hide: Data accuracy, engine coverage, prompt caps, and the real bill after add-ons.
• The top picks: HubSpot for a free, CRM-connected start, Mentions for agencies and white-label reporting.
• Best for creating content: Surfer writes and optimizes for Google and AI in one draft.
• Deep pockets, specific needs: Scrunch tracks crawler traffic, Brand Radar researches any brand, AirOps runs content workflows.

The 7 best AEO tools in 2026HubSpot AEOA natural AEO tool for HubSpot usersHubSpotHubSpotHubSpotHubSpotHubSpotSpecifications

• Tracking depth-4/5
• Setup time-4.5/5
• Insight clarity-4.5/5
• Ease of setup and use-4/5
• Value for money-4.5/5

Pros

• Free 28-day trial
• Recommendations come as a priority task list
• Pulls real buyer prompts from your CRM
• Affordable plans
• Tracks which AI sent each lead

Cons

• CRM prompts locked to Pro/Enterprise plans
• Standalone caps at 25 prompts monthly
• Still in beta; content tools unfinished

HubSpot launched its AEO tool in April 2026, built on tech from xFunnel, an Israeli startup it acquired in late 2025.

The pitch is simple. Track how ChatGPT, Gemini, and Perplexity describe your brand, then tell you what to fix. For a first version, it handles the tracking well. The fixing part is where it gets more interesting, and where the limits start to show.

Why I picked HubSpot AEO

A lot of AEO tools stop at the scoreboard. Here's your visibility score, now good luck. HubSpot goes one step past that, and that step is why it earns a spot here.

Its recommendations come with a priority and a status.

Create a product page on this topic. Get into these two Reddit threads. Write this listicle.

It reads like a task list instead of a chart, which is rarer in this category than you'd think.

The CRM connection is the other real edge. On the Marketing Hub Pro or Enterprise plan, HubSpot builds your prompt list from actual sales calls, support tickets, and site searches, so you're tracking questions your buyers really ask instead of ones you assumed they would. If you already run your business inside HubSpot, that's a head start no standalone tool can match.

However, that CRM trick is locked to Pro and Enterprise. Buy the $50 standalone plan, and you lose it, which means the cheapest version is also the least clever one. Plus, it's still in beta.

The content-creation features aren't fully out, so for now it tells you what to write without helping you write it. Handy, but not the finished product HubSpot is clearly building toward.

Best features

• Recommendations tagged with priority and status
• AI referral attribution built into HubSpot's source tracking
• A free AEO Grader for a quick first look, no signup needed
• CRM-powered prompts on Pro and Enterprise plans, pulled from real buyer conversations
• Citation breakdown by content type (blog, listicle, product page, homepage) and by owned, earned, or peer

Pricing

$50/month standalone, or $45/ month billed annually. It's free inside Marketing Hub Pro and Enterprise, so if you already pay for either, the AEO tool adds nothing to your bill.

MentionsAdvanced AEO tools for agencies MentionsGalleryMentionsMentionsMentionsSpecifications

• Tracking depth-4.5/5
• Setup time-4.5/5
• Insight clarity-4/5
• Ease of setup and use-4.5/5
• Value for money-4/5

Pros

• Fastest, most automated setup I tested
• Tracks eight engines, most on this list
• White-label reports built for client work
• Ties AI visibility to real traffic

Cons

• Cheapest plan tracks only three engines
• No free plan; starts at $49
• 25 prompts on entry plan
• Recommendations lighter than a full platform

Setting up most AEO tools feels like a chore. You pick competitors, write prompts, choose engines, and hope you guessed right. Mentions does almost all of that for you.

You hand it your domain. It scrapes your site, builds a "brandbook" of your offerings, pain points, and competitors, then drafts a starter set of prompts based on what it found. A minute of clicking and you're looking at a live dashboard. Of everything I tested, nothing got me to real data faster.

Why I picked Mentions

Two things pushed Mentions onto this list: how much it watches, and who it's built for.

On coverage, it tracks eight engines instead of the usual three. ChatGPT, Perplexity, Claude, Grok, Gemini, DeepSeek, Llama, and Google's AI Overview. If your buyers are asking Claude or Grok and your tool only checks ChatGPT, you're seeing half the picture. Mentions sees the whole thing.

The other half is reporting. This is the tool I'd actually hand a client. The monthly reports are clean, shareable, and on the Agency plan, fully white-labeled with your own domain and logo. There's a pitch workspace for winning new business and unlimited client sites. If you sell AEO as a service, that beats another pretty chart.

Underneath, the tracking holds up. Visibility, sentiment, and share of voice, switchable in a click. A per-prompt view that shows the exact AI answer and the sources it pulled from. A Sources page that tells you whether the pages AI scrapes actually mention you, which is the question that matters. And an AI traffic view that connects visibility back to real visits, so you can prove it drove something.

Now the catches. That "every model" promise has fine print. The $49 Starter plan tracks only three engines. You need the $99 Pro plan before "all LLMs" really means all of them. Plus, 25 prompts on the entry plan run out quickly once you get serious.

One more thing worth saying plainly. Mentions watches and advises; it doesn't act. It hands you a tidy Kanban board of fixes (write about pricing, chase this site for a mention), but you're still the one writing and publishing. That's fine, as long as you know you're buying a sharp scoreboard, not a content team.

Best features

• Switch between visibility, sentiment, and share of voice in a click
• Tracks eight AI engines, including Claude, Grok, and DeepSeek
• Kanban insights board that prioritizes your next fixes
• Sources page showing whether the pages AI scrapes actually mention you
• White-label monthly reports with your own domain and logo (Agency plan)
• AI traffic analytics tying model visibility back to real site visits
• Plain-English agent that answers questions and builds reports on demand

Pricing

Mentions runs four flat plans: Starter at $49 a month (25 prompts, 3 engines), Pro at $99 (50 prompts, all engines), Business at $199 (100 prompts), and Agency at $399, which adds white-labeling, a pitch workspace, and unlimited client sites.

Semrush AI Visibility ToolkitDetailed AI visibility reports SEMrushSEMrushSEMrushSEMrushSEMrushSpecifications

• Tracking depth-4.5/5
• Setup time-4/5
• Insight clarity-4/5
• Ease of setup and use-4/5
• Value for money-3.5/5

Pros

• Real sentiment and perception analysis
• AI data sits beside your SEO
• Daily prompt tracking, ChatGPT and Google

Cons

• Costs stack fast with seats/domains
• No real free trial

Semrush joined the AEO party last year and has now launched a scaled-up version of the AI Visibility Index in June 2026. It's less a standalone tracker than an AI layer bolted onto the suite you already know.

The good thing here is that your AI visibility sits right next to your rankings, backlinks, and site audits. What sets it apart from the pure trackers is how hard it digs into why AI describes your brand the way it does, not just whether it mentions you at all.

Why I picked Semrush AI Visibility Toolkit

Plenty of tools tell you whether AI mentions you. Semrush is the one that's actually good at explaining why.

Its perception and "key business drivers" reports break down the specific trust factors AI attaches to your brand (a feature that works without a subscription, say, or a product that runs quietly) and show where a competitor owns a factor you don't.

So instead of "our visibility is low," you get "we lose a whole cluster of prompts because rivals own the access-control angle and our pages never mention it." One is a number. The other is a task you can hand to a writer.

Because it lives inside Semrush, the AI data sits next to everything else. You can jump from an AI gap to the keyword behind it, check whether LLMs can even crawl the page (there's a dedicated AI-readiness site audit), and add any prompt to daily tracking across ChatGPT and Google's AI answers in one click.

Now the trade-offs.

The pricing looks friendlier than it plays out. $99 a month sounds fine until you notice it covers one domain, one user, and 25 prompts. Add a teammate and that's another license. Add domains or prompts and the real bill climbs toward a few hundred, sometimes well past a thousand. Plus, there's no proper free trial, only demo reports you can browse.

Like most of this list, it watches and advises. It will tell you what to fix and even hands you AI-generated strategy ideas, but you're still the one writing the content.

Best features

• AI visibility score out of 100, tracked over time
• Daily prompt tracking across ChatGPT and Google's AI answers
• Competitor gap analysis that sorts by the topics you're missing
• Questions tab surfacing real user queries for content ideas
• "Key business drivers" showing which trust factors rivals own
• AI-readiness site audit that flags issues blocking LLM crawlers

Pricing

The AI Visibility Toolkit is $99 a month per domain (billed annually), which buys one domain, one user, and 25 tracked prompts across five engines, plus the AI-readiness site audit.

If you want AI visibility bundled with the full SEO suite, Semrush One starts around $199. There's no standard free trial, though you can browse demo reports first.

SurferPurpose built AEO tool for writers and content teams SurferSurferSurferSurferSurferSpecifications

• Tracking depth-4/5
• Setup time-4/5
• Insight clarity-4.5/5
• Ease of setup and use-4.5/5
• Value for money-3.5/5

Pros

• One draft scored for both Google and AI
• Mention gap turns visibility into outreach
• Tracks five engines, including Claude
• Polished and easy for writers to use

Cons

• AI Tracker is a paid add-on
• $95 add-on for 25 prompts
• AI drafts still need real editing
• Keyword and topical tools feel lighter
• Over-optimizing the score reads robotic

Surfer is a content editor first and an AI tracker second, and its whole idea is that you write one article and optimize it for Google and AI search at the same time, in the same document, against two live scores climbing side by side.

Why I picked Surfer

Surfer's content editor is the best reason to use it, and it's the part that quietly solved the problem this whole list keeps circling.

As you write, an SEO score (how well you match what's ranking in Google) and an AI search score (how well you cover what AI answers pull from) keeps working.

Here, a checklist nudges you to answer the main question in your first 100 words, because Surfer's own data says that's where a large share of AI citations come from. Plus, Surfy, the built-in assistant, already knows your article, your guidelines, and your voice, so its rewrites don't read like generic chatbot filler.

Then there's the AI Tracker, which is where Surfer stops being unique and becomes merely good. It watches your brand across five engines (ChatGPT, Claude, Perplexity, Gemini, and Google's AI Overviews), scrapes the actual answers a real user sees rather than pinging an API, and reads sentiment.

Surfer also added a one-click jump from that gap straight into the editor to write the fix, which no rival quite matches.

But the content engine has limits too. The AI draft gets you most of the way, but it does need real editing. Lean too hard on the content score and your writing turns robotic. The built-in keyword and topical tools are lighter than Ahrefs or Semrush.

Still, if your actual job is producing content that has to win in both Google and AI, Surfer is the one tool here that helps you write it, not just grade it.

Best features

• Topical map that visualizes your coverage and the gaps
• An LLM-optimized content format built for AI citations
• AI Tracker across five engines, with sentiment on higher tiers
• Mention gap showing the exact sites competitors own and you don't
• Google Docs and WordPress integrations writers actually keep open

Pricing

Surfer's core plans run from about $99/ month for Essential ($79 annual, 30 content-editor articles) up to $219 for Scale, with a custom Enterprise tier above that.

The AI Tracker is a separate add-on at around $95 a month for 25 prompts, and the full five-engine, daily-refresh, with-sentiment version sits on the higher tiers.

Ahrefs Brand RadarInstant competitor and market research AhrefsAhrefsAhrefsAhrefsAhrefsAhrefsSpecifications

• Tracking depth-4/5
• Setup time-4.5/5
• Insight clarity-3.5/5
• Ease of setup and use-3.5/5
• Value for money-2.5/5

Pros

• Prompts built from real search data
• Zero setup, instant data on any brand
• Adds web, Reddit, YouTube, and TikTok tracking
• Research any competitor without owning it
• Free demo before you pay

Cons

• No Claude or Grok tracking
• Lacks sentiment analysis
• Costly plans

Here's the dirty secret of AEO trackers. Nobody actually knows what people type into ChatGPT, because that data is locked away. So most tools invent prompts that sound plausible and hope they match reality.

Ahrefs took a different route. Brand Radar builds its questions from its real keyword database, the actual things people search on Google, then runs those through the AI engines. It's still a proxy, and Ahrefs says so plainly. But it's a proxy built on real search behavior instead of pure guesswork.

Why I picked Ahrefs Brand Radar

Type in any brand, yours, a competitor's, or one from an industry you've never touched, and Brand Radar shows its AI visibility instantly. The reach is wide. In one place, you see how six AI engines answer, which URLs they cite, how your branded search volume is trending, who mentions you across the web, and which Reddit threads, YouTube videos, and TikToks are shaping the conversation.

Then there's the data. Where other trackers guess at prompts, Brand Radar pulls from more than 200 million real search queries and weights everything by actual search volume. A mention on a query thousands of people search counts for more than one nobody asks. That grounding is genuinely different.

Now the honest part, because there's plenty of it.

Start with what it can't see. Six engines sounds broad until you notice Claude and Grok aren't on the list. If your buyers use either, that's a blind spot. There's also no sentiment analysis, so you learn whether AI mentions you, not how it describes you. Every cheaper tool here does that much.

The data grounding cuts both ways, too. Google searches aren't AI prompts, and independent tests have flagged accuracy gaps (one found three ChatGPT mentions where there were actually over a hundred). So, treat the share-of-voice numbers as a directional read, not gospel.

It's a research tool, full stop. It shows you the map, then hands the walking back to you. Ahrefs sells a separate agent for this called “Agent A”.

Best features

• Zero setup; the data is already collected and waiting
• Free demo queries so you can try before buying
• Adds web mentions, Reddit, YouTube, and TikTok visibility
• Analyze any brand, competitor, or whole industry instantly
• Prompts built from real Google search data, not synthetic guesses
• Impressions and share of voice weighted by actual search volume

Pricing

Brand Radar is an add-on, not a standalone. Each AI engine costs $199 a month, or $699 for all six bundled, and that sits on top of a required Ahrefs base plan (from $129 a month for Lite).

ScrunchAEO tools for technical and enterprise teams scrunchscrunchscrunchscrunchSpecifications

• Tracking depth-4.5/5
• Setup time-3.5/5
• Insight clarity-4/5
• Ease of setup and use-3.5/5
• Value for money-3/5

Pros

• Tracks the AI bots crawling your site
• Ties AI citations to real conversions
• Both sides: answers and crawler traffic
• Sentiment and citation analysis included
• Enterprise-grade security and integrations

Cons

• Best features locked to Enterprise
• Only four engines on entry
• Too complex for small teams

Here's a question no other tool on this list really answers. When an AI crawler visits your site, what does it actually see, and does that visit ever turn into a customer? Scrunch is built around that question.

It tracks the bots hitting your pages, ties their visits to real referral traffic and conversions, and can even serve those bots a cleaner version of the page. It watches the plumbing of AI search, and that makes it the most technical tool here and the most enterprise.

Why I picked Scrunch

Scrunch connects three things other tools keep separate. Where you show up in AI answers. Which AI crawlers actually hit your site and which pages they read. Plus, through a GA4 hookup, what referral traffic and conversions those citations drove.

So when a comparison page pulls a handful of Perplexity sessions that convert well while a blog post pulls more ChatGPT traffic that converts nothing, you know exactly where to push.

Then there's AXP, its most distinctive bet and the one that needs the most caveats. It sits at your CDN and serves AI bots a stripped-down, machine-readable version of a page (Scrunch's own example crushes a pricing page from roughly 124,000 tokens to about 1,300).

A cleaner page is easier for a model to parse and cite accurately. It's genuinely forward-looking. It's also Enterprise-only, still in limited release, and needs CDN-level setup, so treat it as a reason to keep an eye on Scrunch, not a button you'll press next week.

Best features

• Agent traffic analytics showing which AI bots crawl you
• GA4 integration that ties AI citations to conversions
• Citation drill-down to find the exact pages for outreach
• Sitemaps view that scores every page for AI performance
• AXP serves AI bots a compressed, machine-readable page (Enterprise)
• Enterprise-grade security: SOC 2, SSO, and RBAC

Pricing

Scrunch starts at $250 a month for the Core plan, covering 125 prompts across four engines (ChatGPT, Perplexity, Google's AI, and Copilot).

Higher tiers add more prompts and multi-client support, while the headline features- broader engine coverage, the AXP delivery layer, full agent-traffic analytics, and API access- sit behind a demo-gated Enterprise plan.

AirOpsAEO tool built for running content workflows end to end airopsairopsairopsairopsairopsSpecifications

• Tracking depth- 4.5/5
• Setup time-2.5/5
• Insight clarity-3.5/5
• Ease of setup and use-3/5
• Value for money-3/5

Pros

• Closes the loop from insight to measurement
• Quill agent turns data into content campaigns
• Tracks sentiment and third-party citation sources
• Playbooks run fixes across hundreds of pages
• Publishes finished content straight to your CMS

Cons

• Steep learning curve, expects SEO expertise
• Setup measured in weeks, not minutes
• Overkill for solo or small teams

AirOps is a full content operations platform that happens to do AEO. It watches how AI describes your brand. But then it wants you to build playbooks, run them across hundreds of pages, refresh stale posts, and push the finished article straight to your CMS.

That range is both the draw and the warning. AirOps does more than almost anything else on this list. It also asks more of you than almost anything else on this list.

Why I picked AirOps

AirOps is the only tool here that will actually build and publish the fix for you. Everything else on this list points at the problem and leaves the work to you. AirOps hands you the machinery.

Here's how that plays out. You notice AI keeps describing your app as slow, or skipping you on privacy questions. You open Quill, the platform's AI agent, and just talk to it. It pulls the sentiment data, finds the prompts driving the problem, and helps you spin up a page-refresh campaign. The campaign runs, publishes, and then tracks whether your citations actually climbed. Insight, action, and proof, in one place.

The tracking underneath is deep, too. It follows sentiment, so not just whether you show up but how you get described. It maps which third-party sites (Reddit, YouTube, LinkedIn) are shaping your answers. And its Page360 view stacks AI search, SEO, and citations together, so you can catch a page that gets crawled constantly but never cited.

So why isn't it near the top of this list? Because all that power comes with a tax.

AirOps expects you to already know SEO. It hands you a workflow builder, not a wizard, and if you show up without a plan, you'll sit staring at a blank playbook. Quill softens that, but setup still runs in weeks, not minutes.

Then there's the bill. AirOps charges by the task, so every AI step in every workflow costs you, and a single article can burn hundreds of tasks. Worse, the paid plans leap from about $200 a month to about $2,000 with almost nothing in between. Outgrow the cheap plan, and you don't take a step up; you have to climb a mountain.

For an agency refreshing hundreds of client URLs, that math can work. For a solo marketer or a small team, AirOps is a Formula 1 car booked for the grocery run.

Best features

• Sentiment tracking, how AI describes you and not just whether
• Quill, an AI agent you chat with to launch campaigns
• Playbooks that apply one fix across hundreds of pages
• Third-party citation tracking across Reddit, YouTube, and LinkedIn
• Native CMS publishing to WordPress, Webflow, Sanity, and more
• Page360 view stacking AI search, SEO, and citation data in one screen

Pricing

AirOps starts with a free single-user tier, then the paid plans jump from a Solo plan around $200 a month to a Pro plan around $2,000, with almost nothing in between. Billing is usage-based, so you pay per task (each AI step counts), and the paid tiers are quote-based.

What are AEO tools?

AEO stands for Answer Engine Optimization. It's the work of getting your business mentioned, cited, or recommended when someone asks ChatGPT, Gemini, Perplexity, or Google's AI features a question you could answer.

AEO tools are the software that supports that work. These tools generally do the following tasks:

• Track your visibility: Show you whether, and how often, AI engines mention your brand, where you rank in the answer, and how you stack up against competitors.
• Find the questions people ask: Surface the real prompts your customers are typing into AI, so you know what your content needs to answer.
• Show which sources AI trusts: Reveal the websites, pages, and threads AI pulls from when it answers, so you know where to show up, not just on your own site.
• Create and optimize the content: Help you write and structure pages, so both Google and AI are more likely to pick them.

Importantly, Google has publicly stated that optimizing for its AI search features is still just SEO. There's no secret AI algorithm that a special tool unlocks.

So treat every tool in this guide as what it is: a way to do the fundamentals faster, see them more clearly, or prove they're working.

How I tested these AEO tools

I used each of these AEO tools the way you would on a real project, then scored what it was like to live with.

These tools don't all do the same job. Some track your visibility, some research questions, some write content, some watch your crawler traffic. So I didn't hold a tracker to the same bar as a content platform. I judged each one on how well it does the thing it's built to do.

Every tool got a score out of five in five areas:

• Tracking depth: How much the tool actually captures. Depending on what it's for, that means prompts, brand mentions, citations, competitors, content performance, historical trends, and how many AI engines it covers. More depth, higher score.
• Setup time: How long from signing up to seeing something useful. Tools that got me to real data fast scored well. Tools that made me sit through long onboarding or heavy manual setup scored lower.
• Insight clarity: How easy the findings are to read. Could I look at the dashboard, the reports, and the recommendations and know what to do next? Or did I have to decode it first?
• Ease of setup and use: How the tool feels once you're past setup. Getting around, managing projects, running everyday tasks, pulling reports, and how steep the learning curve is.
• Value for money: What you get for the price. I looked at the limits, the features, and who each tool is for. An expensive enterprise platform isn't bad value if it earns its price for a big team, and a cheap tool isn't good value if the entry plan can't do anything useful.

The overall score is the average of those five. I also weighed what each tool is for, so a specialist tool didn't lose points for skipping features it was never trying to have.

What to look for in an AEO tool?

The marketing pages all look clean. The complaints show up later, once you're paying. Here's where these tools actually crack, and what to check before you input your credit card.

• Accuracy you can trust: This is the number one gripe by a wide margin. The data often doesn't match reality. Ask how the tool collects data (does it read the real answer a user sees, or just ping an API?) and how often it re-runs each prompt. Treat any single-snapshot score as a rough estimate, not gospel, and don't panic over small week-to-week swings.
• Real prompts, not made-up ones: Plenty of tools invent plausible-sounding queries and call it monitoring. That's sampling with a confidence problem. Check that you can add your own prompts, and better still, feed in the real questions your buyers ask (from your search console, sales calls, or support tickets).
• Engines your buyers actually use: Most tools quietly track ChatGPT and little else. If your customers lean on Gemini, Claude, or Grok, confirm coverage before you pay, and confirm it's on the plan you're buying.
• A score you can act on: A lot of dashboards look impressive and tell you nothing to do next. If the tool can't show you the why behind the number (the sentiment, the prompts, the sources feeding the answer) and a clear next step, you're paying to watch a chart move.
• The real monthly bill: The headline price is the tease. Watch the prompt cap (10 to 25 runs out fast), per-seat fees, per-domain fees, usage-based task billing, and add-ons.
• Tracker or doer, and does it fit your stack: Decide whether you need something that watches or something that writes, because few tools do both well. And favor one that plugs into what you already run.

Buy for the job in front of you. The worst subscription is the impressive one you never had a use for.

Quick AEO tool selection guide

Short on time? Find your situation, grab the tool, skip the rest.

• Already live in HubSpot? HubSpot AEO gives CRM-fed tracking and a priority task list, cheaply.
• Need content written and shipped at scale? AirOps builds, runs, and publishes it, if you know SEO.
• Running an agency? Mentions sets up fastest, tracks eight engines, and white-labels your client reports.
• Already pay for Ahrefs? Brand Radar adds AI research built on real search data (no sentiment).
• Already on Semrush? Its toolkit explains why AI describes you that way, besides your SEO.
• Technical or enterprise team? Scrunch tracks AI crawler traffic and ties citations to real conversions.
• Want to write for Google and AI at once? Surfer optimizes one draft against two live scores.
• On the tightest budget? Start with a free checker, prove AI search matters, then buy.

FAQsWhat are the best AEO tools for ecommerce?

For ecommerce, Semrush and Mentions track product-level visibility well, and Surfer helps optimize product and category pages. But the real groundwork is clean product schema (price, reviews, availability) so AI can read and recommend your catalog. Pair a tracker with solid structured data.

Are AEO and GEO the same thing?

Not exactly, though people now use them interchangeably. AEO (Answer Engine Optimization) is about becoming the direct answer AI gives. GEO (Generative Engine Optimization) is about getting cited inside AI-generated responses. They overlap heavily and share the same fundamentals, so most tools cover both under one label.

Will AEO replace SEO?

No. AEO adds to SEO, it doesn't replace it. Google has said optimizing for its AI features is still just SEO, with the same index and fundamentals. Strong SEO is what gets you into AI answers in the first place. Treat AEO as a layer, not a swap.

Do I need a paid AEO tool, or is a free one enough?

A free checker is enough to start. It tells you whether AI mentions you and where the gaps are. Once you need ongoing tracking, competitor comparison, and sentiment over time, a paid tool earns its place. Start free, then upgrade when you have something to act on.

How do you measure AI visibility?

AI visibility is measured by mentions, citations, share of voice, and sentiment, not rankings or clicks. Mentions count how often AI names you, citations track when your page is the source, and sentiment shows how you're described. Most tools in this guide report all four.

Why doesn't my AEO tool's data match what I see in ChatGPT?

Because AI answers change constantly. The same question returns different results by phrasing, location, model, and day. Most tools sample a set of prompts and take a snapshot, so their numbers are estimates, not exact counts. Treat the trend as directional and don't over-read small swings.

Which AI engines should an AEO tool cover?

Cover the engines your buyers actually use. ChatGPT, Google's AI Overviews, and AI Mode matter for almost everyone. Add Perplexity, Gemini, Claude, and Grok depending on your audience. Check coverage before you pay, since many tools track only ChatGPT and charge extra for the rest.

How much do AEO tools cost?

AEO tools range widely. Free checkers get you started, entry plans run about $49 to $99 a month, mid-tier tools land around $200 to $700, and enterprise platforms climb past $2,000. Watch for prompt caps, per-seat fees, and add-ons that quietly raise the real bill.

Here’s what CNET readers predict Apple will do this fall.

As AI demand accelerates infrastructure growth, the binding constraint is shifting from technical capacity to local planning risk.

That risk is inseparable from how well developers treat the people living next to what they build.

Angela Rayner approved a hyperscale data center at Woodlands Park in Buckinghamshire in July 2025, overturning a local council that had rejected it twice.

Six months on, the government's own lawyers admitted the approval rested on a "serious logical error," because the planning inspector had waived a full Environmental Impact Assessment and several promised mitigation measures were never secured.

Whatever the legal challenge decides, the lesson is already visible. This wasn't an engineering failure or a financing failure. It was a planning failure, and it's the kind data center developers are going to keep running into.

Capacity, capital, and political appetite for AI infrastructure aren't the constraint anymore. Getting a shovel in the ground without a council reversing course, or a court quashing the decision months later, increasingly is.

How fast this is moving

Ofgem's own figures put a number on how fast this is moving. Around 140 data centers in the UK are seeking grid connections worth roughly 50 gigawatts, more than the country's winter peak demand of 45 gigawatts.

In Japan, residents in Inzai and the neighboring city of Shiroi have filed lawsuits over facilities built meters from their homes, arguing nobody told them what was coming until construction had already started.

In Australia, opposition has grown fast enough that state parliaments are now holding inquiries into how planning approvals were reached. These are planning system failures, and they’re arriving faster than the systems built to handle them were designed for.

That's not the same as saying developers are acting in bad faith. Grid connection queues in parts of Japan now stretch five to ten years, and several UK regions aren't far behind. Somewhere in that queue right now is a developer with a signed lease and a hired crew, with nothing to build on for years.

Capital gets committed against a five-year AI roadmap and then sits stalled behind infrastructure queues that predate the AI boom by decades. There's a real competitive cost too: a country that builds slower loses the workloads, jobs, and tax revenue to one that doesn't.

Understanding that pressure matters, because it's a big part of why planning gets treated as an obstacle to clear quickly rather than a process to build around.

Local resistance

None of that makes local resistance irrational. In June, the US energy secretary Chris Wright told data center backers at an industry conference to push back harder against critics, calling concerns about water and power use "overblown."

Diesel backup generators running test cycles at odd hours, or a permanent hum next to a family home, aren't things people object to for the sake of it.

Water is the sharpest version of the problem: Thames Water estimates a single large data center can draw up to 19 million liters a day, and the UN's Special Rapporteur on the human right to water has called for a moratorium on data center expansion over exactly this kind of pressure.

A separate scheme at Wapseys Wood in Buckinghamshire would run entirely on gas turbines next to two primary schools and a hospital. National strategy and local reality are being asked to occupy the same site, and only one of them has to live there.

Most of the friction traces back to sequencing. Consultation tends to start after a site is chosen and a design is fixed, which means residents are reacting to decisions instead of shaping them, and that timing alone tells a community what its input is worth. Woodlands Park already showed what that looks like. An environmental assessment waived. Mitigation promised and never secured.

Reporting thin enough that it took a legal letter to force the numbers into the open. Local benefit gets described in broad terms, jobs and regional investment, rather than anything a resident could actually measure or hold anyone to. Construction disruption is consistently underestimated in planning submissions and then lived with daily by people who never asked for a data center next door.

Planning systems increasingly react to disputes rather than shape outcomes before they harden. That's especially true where fast-track routes and national growth zones have been layered on top of existing local plans.

The actual fix

The actual fix is reordering it, not slowing everything down. Ireland's grid connection regime, introduced in late 2025, requires new data center developments to build their own generation or storage capacity and hit renewable thresholds near 80 per cent over time, rather than draw on shared grid capacity for free.

That forces developers to carry a cost most markets still push onto residents and ratepayers. Local benefit needs the same treatment: a fixed share of revenue or tax contribution routed to local government on a schedule, not a discretionary grant pot handed over once objections have already surfaced.

Engagement has to start before a planning application is drafted, not after, so siting and noise mitigation can actually be shaped by what residents raise rather than defended against it. And liaison can't end when construction does. A facility runs for decades inside a community, and a named contact five years after opening matters more than one consultation event before it.

Data centers are core national infrastructure now, not a niche technical build on an industrial estate. But infrastructure that treats the people living around it as an afterthought doesn't get delivered on schedule, whatever the national strategy document says. Local legitimacy sits alongside grid capacity and water availability as a delivery constraint in its own right, not somewhere beneath them.

Developers who design for that early will move through planning faster and face less resistance once they're running. The ones who don't will keep finding out, one quashed permission and one lawsuit at a time, that a planning problem left alone doesn't become a technology problem.

It becomes a more expensive planning problem with worse PR.

We've tested, reviewed, and ranked, the best web hosting services.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

Robot lawn mowers are emerging as one of the most exciting smart-home categories. The concept has been around for a while, but in the past two years we've seen advances in technology that are making them more helpful and more user-friendly than ever. Finally, the barriers to entry are coming down, and lawnbots are starting to look like a feasible option for even the least tech-minded amongst us.

To find out more about what's going on in this growing space, I spoke to KK Yin, marketing director at major lawnbot brand Mammotion. We chatted about the most impactful shifts in lawnbot technology, the hottest trends, and the hurdles the industry still needs to overcome.

Cutting-edge edge cutting

For KK, the big feature that hasn't quite been nailed yet is edge-cutting. Even if you have a great robot mower that does a stellar job trimming your grass, its usefulness is limited if you then have to get out a trimmer to take care of the edges. "Edge trimming remains one of the biggest challenges, especially along walls," says KK. "The key is balancing cutting performance with safety."

The key is balancing cutting performance with safety

If you're trimming along the edge of a flat path, it's relatively easy — certain mowers are designed to straddle the path edge, with the blades underneath trimming the grass at the boundary line. However, that doesn't work when trying to trim next to a wall. "The lawnmower can get as close as possible, but the cutting disc being set back from the mower's edge to maintain a safe clearance prevents the mower from cutting close enough to walls," KK continues.

Lots of brands have been working on a solution, and it typically takes the form of an edge-trimming module that attaches to the side of the bot, with smaller blades in a protective casing.

An edge-cutting attachment on a Roborock lawnbot (Image credit: Future)

On the Luba Mini is an edge-cutting disc that can trim as close as 5cm (1.97 inches) from a wall — a distance KK admits might not be close enough. "Five centimeters is not very far from the wall, but still some users want us to make it perfect so it can cut maybe even down to one centimeter," KK says.

"We're exploring more flexible solutions," he continues. "For example, when mowing alongside a wall, the cutting disc could potentially extend slightly out of the mower's body for a cleaner edge cut. In open areas or during ride-on-the-edge mowing, it would retract to maintain a safe clearance beneath the mower. So that is the direction and also the challenge in the future."

Tackling tricky terrain

Another major lawnbot trend that's been emerging over the past year or so is all-wheel drive (AWD). This helps lawnbots tackle tricky or uneven terrain, as having all the wheels individually powered means that if one or two end up off the ground, or struggling to get traction, the others can help the robot escape without requiring manual intervention.

AWD can also help the lawnbot retain its grip when scaling especially steep lawns. KK notes that in the European market, and Germany especially, lots of houses stand on slopes, where this boosted climbing ability comes in handy.

All-wheel drive is usually indicated by an 'AWD' on the product's name, and it appears in Mammotion's advanced Luba line — you can see the LUBA 3 AWD scaling an astroturf mountain in the video above.

The brand has also been looking at upgrading its bots' climbing ability and mowing speed, and is keen to figure out what the next advancement might look like. "We can see some other companies are exploring other things like independent steering," says KK. "Maybe that is a good solution for difficult terrain. So we are also thinking about how to combine this technology."

Better tech in smaller bots

Initially, Mammotion kicked off its lawnbot journey with its premium Luba series. Later on, the Yuka Mini arrived — a smaller, more affordable bot that's designed for yards 500 to 800 square meters in size. It proved to be the big winner in Europe during the recent Amazon Prime Day sales.

"I think when we started, our premium product line helped us build trust in our technology — things like the all-wheel drive and the positioning system and the efficiency. So customers believed our brand could bring some of that technology to an entry-level lawnmower," muses KK.

The compact Yuka Mini is a good fit for smaller, European yards (Image credit: Future)

The Yuka Mini is a good fit for European yards, which on average are 300 to 500 square meters in size. "Many robot lawnmower brands spend a lot and have a different product strategy for European regions, because many of the users have already accepted this innovative product for their lawn care. For example, they know about the positioning systems and how all-wheel drive compares with rear-wheel drive. They want their robot lawnmower to do more things — not only mowing, but also edge trimming."

In the US, where the average lawn size is much larger than in Europe, the focus is a little different. "I think most US users still prefer a ride-on or traditional lawnmower. I think they're still looking for an efficient robot mower that can cover their large lawn," says KK. "The US is at a different stage, and what customers require is different.

No-fuss installation

Navigation tech is perhaps the area that has moved on most quickly and has had the most impact on lawnbot usability. It was only a few years ago that boundary wires were required; now they're all but obsolete.

Initially, the technology to take over was RTK. This is satellite-based positioning, and it requires a separate RTK receiver to improve the accuracy of the satellite data. Not everyone wanted to set up a receiver in their yard, leading to a shift to centralized receivers that could serve massive areas at once. Mammotion calls this NetRTK.

Either way, the system relies on satellites, which means it has some limitations in yards where there is coverage from trees or buildings. Most recently, we've started to see LiDAR being used. This is the navigation approach used in robot vacuums, and just like with RTK, there are pros and cons when used in lawnbots. LiDAR works by bouncing light off objects to build up a picture of its surroundings, which is great when there are objects around but less useful in wide, open spaces with nothing to bounce off.

We're seeing a shift away from having to install an RTK receiver in your yard (Image credit: Future)

"You know, like we always say, maybe laziness is the best way to improve the technology!" says KK. "For our users, if they're happy with the intelligence of the robot lawnmower, then they want an easier installation, or even no installation at all. So that's why they really like our NetRTK, because they don't need to install an antenna. Or some users will choose LiDAR, which doesn't need an antenna at all."

The real shift is less to do with using one specific navigation approach over another, and more about moving to a situation where multiple navigation technologies can work alongside one another.

"We want to think about the final result of the positioning system," says KK. "Many users don't want to compare which positioning method is good. They just use it; that's enough. They don't need to know more about it. So that's why we came up with Tri-Fusion, which combines the advantages of RTK and LiDAR and even AI vision. We believe this can cover all environments for a residential user."

Users want an easier installation, or even no installation at all

A few robot mower companies have started experimenting with similar concepts, where the bot can swap between different approaches depending on the situation. If satellite signal drops out under some trees, LiDAR can swoop in to save the day and get the lawnbot back on track, for instance. Cameras (often backed by AI) form the third pillar, enabling the bot to recognize objects and understand if it should mow close to them or steer clear.

There are problems still to solve in the world of robot mowers, but let's not forget that just a few short years ago, we were still painstakingly laying boundary wires and watching the lawnbot bounce haphazardly between them. Now, the issue is whether a bot can trim quite close enough to the edge of a wall, and if it's smart enough to tell a soccer ball from a lounger. We've come a long way.

Brilliant bots

We've tested plenty of impressive lawnbots at TechRadar. Here's a taster of some of our favorites — hit the View details button for a link to our full review for each.

Mammotion LUBA 3 AWD 3000

Read our full review

Mammotion Yuka Mini Robot Lawn Mower

Read our full review

Mammotion YUKA Mini 2 1000 Robot Lawn Mower with LiDAR

Segway Navimow i210E LiDAR Pro

Read our full review

Segway Navimow X3 Series

Read our full review

TerraMow V1000 Robot Lawn Mower

Read our full review

ANTHBOT Genie Smart AI Robot Lawn Mower

Read our full review

Eufy E15 Robot Lawn Mower

Read our full review

Mammotion LUBA 2 AWD

Read our full review

The AI race is usually framed as a contest for chips, models and talent. Increasingly, it is being decided by something more basic: power.

The International Energy Agency (IEA) projects that electricity demand from data centers will double by 2030.

At the same time, many grids are already under strain from increased electrification, ageing infrastructure and long permitting cycles.

This makes the ability to secure reliable electricity on a timeline that matches business plans a decisive factor in choosing where to build an AI data center.

Time to power is becoming a key constraint

That is why “time to power” is quickly becoming the key metric in site selection.

Schneider Electric’s work in this space shows that in multiple regions, grid connection approvals and reinforcement timelines that were once manageable now stretch far beyond normal development cycles.

In energy-constrained hotspots, the process to connect to the grid can extend for years, and in some cases can reach a decade or more. For large AI data center projects, that changes where data center operators build and how they build.

The issue is not the lack of power. The real challenge is that data centers concentrate in specific regions, and those local grids hit capacity limits faster than new infrastructure can be planned, approved and built. When that happens, power availability starts to overtake “location” as the deciding factor.

Transparency is key to attracting investment and ensuring resilience

Power availability also influences where companies are investing when it comes to building AI data centers as it is key to resilience. The most attractive markets are those that offer certainty and transparent information on available grid capacity and reinforcement plans.

Just as importantly, data center operators need confidence that grid connection queues are managed in a way that prioritizes projects that are ready to build, rather than allowing speculative projects to sit in line for years and block capacity.

Resilience has always been fundamental to data centers, but it’s now being redefined. It is no longer only about redundancy inside the facility; it is equally about resilience to the grid environment around it. In many energy-constrained regions, shrinking grid capacity and congestion increase the risk of outages and emergency grid measures.

Hybrid power generation strategies offer an alternative to legacy grid dependence

These pressures are driving many data center developers and owners to consider on-site hybrid energy strategies which combine on-site power generation from renewables with electricity provided by the grid.

Such hybrid power approaches are key for reducing reliance on legacy grid infrastructure and enabling data center operators to generate energy on-site and improve resilience. This also increases energy flexibility by allowing data center operators to change when and how energy is used, stored, or generated in response to demand, prices, or grid conditions.

A concept that is getting more attention in this space is “energy parks”. In simple terms, an energy park brings together multiple power resources, from renewables, thermal power, and storage to on-site generation, and connects them to the grid through a single point. These energy parks may connect to the grid through a single interconnection point, operate as fully islanded systems, or combine both approaches through hybrid architectures.

Because the generation, storage and grid connection are designed together, energy parks can offer a faster or more predictable “time to power” in places where the traditional grid connection queue is a challenge. This could help alleviate the grid connection issue for many data center developers.

A new approach to designing data centers

This is where energy strategy becomes a design strategy. Developers need to plan across three key areas: how to connect, how to operate, and how to scale.

Connecting to the grid is now about getting the earliest possible clarity and avoiding prolonged delays. Operating is about building for an environment where grid constraints, curtailment risk and price shocks can impact business continuity. Scaling is about making sure a site that works at 50 MW does not become constrained at 150 MW because the upstream network cannot grow with the campus.

But none of this can be separated from sustainability. Sustainable AI infrastructure at scale requires electrification strategy and grid modernization: faster grid reinforcement, better planning, and more flexibility on both the supply and demand sides. It also requires closer coordination between developers, grid operators, policymakers and clean energy providers. This will help ensure that digital infrastructure growth is aligned with clean capacity additions and realistic delivery timelines.

For data center developers, the practical takeaway is straightforward: start with power and treat energy as a critical design variable. Build a site selection process that tests grid readiness and timelines before land is locked in, and design an energy strategy that covers connection, resilience and scale from day one.

When it comes to attracting AI infrastructure investment, the winners will be those that can offer speed and certainty on grid connection and reinforcement, and enable credible pathways to clean, resilient power.

Best web hosting services: 60 sites tested to find the 10 fastest, most reliable platforms.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

• Control Resonant lead gameplay designer Sergey Mohov isn't worried about launching the game close to GTA 6
• He says the game's release date has been locked in for a while
• Mohov explains that "It's not really been a big consideration for us"

Remedy Entertainment isn't worried about launching Control Resonant so close to Grand Theft Auto 6.

GTA 6 is the most anticipated game of the year, if not of all time, and is on track to earn over $5 billion by the end of its launch week, so it's understandable if some studios want to avoid releasing their games so close to its launch for fear of being overshadowed.

After Rockstar Games formally announced that its game will arrive on November 19, Playground Games delayed Fable to February 2027 "so it can have the dedicated moment it deserves."

However, games like Marvel's Wolverine and Phantom Blade Zero are still locked in for their Fall releases, so not every studio is worried about competing with GTA 6 or is happy with the amount of breathing room they do have, at least.

Speaking in an interview with TechRadar Gaming about the upcoming release of Control Resonant, lead gameplay designer Sergey Mohov explained that he's looking forward to GTA 6 like everybody else, but doesn't think the game's launch affects Remedy "at all."

"Honestly, first of all, personally, I'm looking forward to GTA as much as everybody else," he said. "I'm a fan of [GTA] Vice City. But I don't think it's affected us at all, to be honest."

Control Resonant launches on September 24, which is eight weeks before GTA 6, so the Rockstar game shouldn't overshadow the former, and Mohov explained that Remedy's release date has been "locked in for a while."

"We expect it to be a very complete experience on day one when it comes out," he said. "We've been planning to release this game in September, for as long as I can remember. It's not really been a big consideration for us."

He added, "Obviously, this is a really exciting year to be a gamer. We're gonna be playing all of these [games] same as everybody else, but hopefully, you pick up, and your readers pick up Control Resonant as well."

Control Resonant will be available for PS5, Xbox Series X, Xbox Series S, and PC.

Did you know TechRadar now has membership?

(Image credit: Future)

Become a TechRadar Insider by simply clicking 'Join Now' at the top of this page. Have a question? Please email membership@techradar.com

Q Acoustics 3040c review: Two-minute review

The 3040c is the smaller of two floorstanding speakers in Q Acoustics' celebrated 3000c range — it’s the fifth and final 3000c model, and it’s arguably the best of the lot.

For £599 or equivalent, depending on where you reside, you'll receive two reasonably compact, elegant and nicely finished cabinets (available in four different finishes) with optimised point-to-bracing inside and a three-strong driver array on the outside.

Two 120mm mid/bass drivers sandwich a 22mm soft-dome tweeter that’s decoupled from the rest of the cabinet — and with some input from a rear-firing bass reflex port, a claimed frequency response of 43Hz - 30khz is the result.

Sensitivity of 88dB means you don’t need all that burly of an amplifier to drive these Q Acoustics speakers effectively.

And when they’re being driven effectively, there’s plenty to like in ultimate terms and even more when you keep the price of these speakers uppermost in your mind. They’re a balanced, poised and expressive listen, good at soundstaging, better at maintaining a convincing tonal balance, and best of all when it comes to detail retrieval.

Some of the best stereo speakers on the market? Yes. OK, they could do with a touch more dynamic headroom and a slightly more assured method of expressing rhythms, but really I’m tempted to ask what more you might realistically hope for when listening to a pair of compact, well-made floorstanders for this money…

(Image credit: Future / Simon Lucas)Q Acoustics 3040c review: Price & release date

• $1,099 / £599 (around AU$1,140, based on UK pricing)
• Released July 30, 2026

The Q Acoustics 3040c passive floorstanding loudspeakers are on sale now, and in the United Kingdom a pair will cost you £599. This translates to €899 in the Eurozone and £1099 in the United States (yes, a little pricier in the States) while prices for other territories, including Australia, are yet to be confirmed.

Competition? You'd really be looking at bookshelf speakers for anything as affordable as this from a trusted name — and let me be clear, Q Acoustics is one such name. So, it's a lot of speaker (and speaker tech) for the price.

(Image credit: Future / Simon Lucas)Q Acoustics 3040c review: Features

• 22mm soft dome tweeter
• 2 x 120mm Continuous Curved Cone mid/bass driver
• 43Hz - 30kHz frequency response
• 88dB sensitivity

No one’s expecting a whole lot of features when it comes to reasonably affordable passive loudspeakers, which is just as well given the Q Acoustics 3040c have just the essentials.

As is the company’s standard operating proactive, though, these few features are completely fit for purpose.

On the inside, Q Acoustics has used its targeted ‘point-to-point’ bracing technology — having identified those areas of the cabinet that are prone to low frequency vibrations, ‘P2P’ reinforcement seems to minimise those vibrations and produce a cleaner, better focused soundstage as a result. It’s also deployed ‘Helmholtz Pressure Equalisers’, a series of tubes that minimise internal pressure; this reduces the build-up of standing waves inside the cabinet and keeps sonic colouration to a minimum as a result.

At the rear of the cabinet there are a couple of very low-profile speaker binding posts, positioned just below a fairly large bass reflex port. Q Acoustics provides foam bungs for the ports — so feel free to experiment with the position of the speakers and the effect of the bungs too.

Up front there are three drivers that are familiar from some of the more expensive model ranges in the Q Acoustics line-up. Here, a decoupled 22mm soft-dome tweeter is positioned between two 120mm mid/bass drivers — these are of the ‘continuous curved cone’ design that made its debut much further up the Q Acoustics loudspeaker range.

‘‘C³’ is designed to combine the low-frequency response of a straight conic cone with the control and clarity more commonly associated with a flared cone design. It’s a straightforwardly impressive design with, at this point in time, nothing to prove.

• Features score: 5 / 5

(Image credit: Future / Simon Lucas)Q Acoustics 3040c review: Sound quality

• Full-scale sound with drive and poise in equal measure
• Great facility with detail retrieval
• Could use greater dynamic impetus

Every audio product reviewed at TechRadar (and there are myriad options) has be put into proper context, of course — and so while I’m going to open this section with a couple of mild-ish criticisms, it’s important to remember that these are relatively affordable floorstanding loudspeakers that are likely to form part of a relatively affordable stereo system. Agreed? Good.

So when playing a copy of Phoebe Bridger’s Lost Boys the 3040c don’t have quite the dynamic potency to track the changes in volume or attack quite as fully as they might. They can do more than allude to changes in intensity, of course, but they don’t put quite as much distance between ‘minimum’ and ‘maximum’ as is absolutely ideal.

And while I’m finding fault, there’s no denying the Q Acoustics don’t quite have the instinctive flow that’s required to fully express a rhythm or manage a tempo. Control of low-frequency activity is never in any doubt, and there’s no suggestion the speakers can drag at the rhythm of a recording — but there’s no denying the 3040c could sound more fluent, especially when asked to play overtly rhythm-based material.

But with this griping out of the way, I can concentrate on all the ways the 3040c is straightforwardly impressive.

It’s a very nicely balanced listen where tonality and frequency response are concerned, for starters. Previous Q Acoustics loudspeakers have tended to err on the warm-ish side of neutral when it comes to tonality, but the 3040c is not at all inclined to stick its oar in — any heat (or lack thereof) in the sound you’re experiencing is there before it reaches the speakers, either in the recording or in the source equipment that’s delivering it.

The speakers are poised and even from the bottom of the frequency range to the top; there’s no overstating of any part and no underplaying, just a smooth and balanced representation. The integration of the drivers is first class, and the point at which handover from the bigger drivers to the tweeter takes place is imperceptible.

As well as the low-frequency punch and substance I mentioned, the Q Acoustics deal with the attack and decay of bass information fairly confidently — and they’re just as adept when it comes to midrange resolution. The 3040c are a detailed listen in every respect, and this manifests itself particularly where voices are concerned. There’s plenty of information about timbre and texture made apparent. At the top end there’s similar alacrity, and more than a hint of bite and shine — but with the substance to ensure treble sounds are never hard or edgy.

Despite the lack of ultimate dynamic headroom, the Q Acoustics do good work with the more minor, but no less significant, dynamics of tonal and harmonic variation. They can identify and reveal even very fleeting and/or subtle occurrences in a recording, and coherently lay them out as part of the overall presentation. If there are details embedded in a recording, the 3040c do their utmost to get them to you.

• Sound quality score: 4.5 / 5

(Image credit: Future / Simon Lucas)Q Acoustics 3040c review: Design

• 942 x 285 x 281mm (HxWxD)
• Magnetically attached grilles
• Choice of four finishes

You have to hand it to Q Acoustics: the 3040c look, and even to an extent feel, like much more expensive loudspeakers than they actually are.

The standard of build and finish is excellent, and even the vinyl wrap that covers the MDF cabinets of my ‘Claro’ walnut review sample is a cut or two above the norm — and I am confident that is also the case if you prefer the ‘pin’ oak, satin white or satin black alternatives.

The 942 x 285 x 281mm (HxWxD) dimensions mean the proportions of the cabinet are harmonious. That ‘width’ measurement includes the small aluminium stabilisers that fit to the back of each cabinet and make it less susceptible to rocking or, heaven help us, toppling. The look of the cabinet is further enhanced by the crisply rendered curves along each horizontal edge.

Q Acoustics supplies magnetic grilles, but for my money you’re better off leaving them in the packaging and enjoying the sight of the driver array instead. The satin nickel brightwork that surrounds them is a confident touch, and another little flourish that suggests the 3040c might cost a fair bit more than they actually do.

• Design score: 5 / 5

(Image credit: Future / Simon Lucas)Q Acoustics 3040c review: Usability & setup

• Attach the stabilisers and spikes
• Connect to an amplifier using appropriate speaker cable
• Toe in just a little towards your listening position

Well, they’re passive loudspeakers and so setting them up and using them is hardly rocket science.

You’ll need to attach the little outrigger stabilisers to the bottom of each cabinet, and then attach the floor spikes. These can be adjusted from the top, so getting the cabinets level is no hardship.

Then, connect the speakers to the amplifier that’s driving them with speaker cable — ideally two identical lengths. And then try to make sure the speakers are toed in, just a fraction, towards the spot from where you do your listening.

• Usability & setup score: 5 / 5

(Image credit: Future / Simon Lucas)Q Acoustics 3040c review: Value

• You'll struggle to find decent floorstanding speakers for this money
• Q Acoustics is a trusted name — and the 3040c are worthy of that heritage

This money won’t even get you on the ladder at some loudspeaker companies, but Q Acoustics first made a name for itself by redefining what ‘entry level’ ought to mean where booth sonic performance and build quality are concerned.

It almost goes without saying that the 3040c does nothing to spoil this reputation — and as a result, it represents prodigious value for money for anyone who wants a fairly big sound from a fairly compact and unarguably affordable floorstander.

• Value score: 5 / 5

(Image credit: Future / Simon Lucas)Should I buy Q Acoustics 3040c?Q Acoustics 3040c scorecard

Attribute

Notes

Score

Features

One cannot expect a plethora of features with passive speakers, but Q Acoustics' proprietary audio architecture is beautifully made and finished

5 / 5

Sound quality

I have minor gripes with dynamic impetus — borderline churlish for this money, but it's my job to notice

4.5 / 5

Design

They look and feel far more expensive than they are

5 / 5

Setup & usability

Simple and easy, and not just because they're passive speakers (although that helps)

5 / 5

Value

This money won't get you on the ladder with most trusted speaker brands. Here, it gets you floorstanders…

5 / 5

Buy them if…

Your room is on the smaller side
These are compact towers, and while they need a bit of space in which to operate they don’t need loads

You want a speaker capable of surviving an upgrade or two
The 3040c will work happily with entry-level amplification, but are able to punch above their weight if you try something a little more capable

You appreciate a bit of stealth design
There’s nothing showy or try-hard about the industrial design language Q Acoustics is dealing in here

(Image credit: Future / Simon Lucas)Don’t buy them if…

You listen to lots of rhythm-centric music
They don’t have two left feet or anything, but the 3040c aren’t the most fluid when it comes to rhythmic expression

You have boisterous pets (or boisterous friends)
Those stabilisers are a good idea, but it doesn’t mean the Q Acoustics are topple-proof…

You like a bit of shock and awe in your music
The 3040c are just fractionally inhibited when it comes to large-scale dynamics

Q Acoustics 3040c review: Also consider

Part of the appeal of the 3040c, in addition to the way they look and sound, is the money Q Acoustics wants in exchange for a pair. It’s genuinely hard to find a pair of floorstanding loudspeakers from a brand anything like as credible as this one for anything like the money — even brands such as DALI, Fyne Audio or Monitor Audio (to name just three) start their floorstanding ranges with models costing several hundred of your chosen currency more than this.

If you're happy to consider bookshelf models however, look to the Dali Sonik 1 or even Cambridge Audio L/R S. But again, they're not floorstanding speakers.

How I tested the Q Acoustics 3040c

I connected the 3040c to a WiiM Amp Pro using a couple of lengths of QED loudspeaker cable, and then connected a preamplified Technics turntable to the WiiM’s single analog input.

I used a Rega CD player, connected to the amplifier’s digital optical input, and also streamed via Bluetooth and Tidal Connect using an Apple iPhone 15 Pro. Which meant I was able to listen to music in lots of different formats, and at lots of different resolutions, more-or-less non-stop for a week or so.

• Read TechRadar's Reviews Guarantee
• First reviewed: July 2026

I asked ChatGPT for one sensible answer, one slightly reckless answer, and a third that combined the best parts of both. The difference was immediate. Instead of giving me the usual safe, predictable advice, it laid out three genuinely distinct approaches — and helped me see the tradeoffs between them.

That small prompt trick solves one of ChatGPT’s most persistent problems: its tendency to give you the most obvious reasonable answer and stop there.

Large language models are very good at identifying common patterns. Ask for the best vacation plan or how to organize your fridge, and you will usually get something perfectly sensible — but also fairly basic. They know what advice usually works, what people typically recommend and what has become accepted wisdom. That makes them useful, but often bland.

The solution is to ask for three kinds of answer: the conventional approach, the unconventional approach, and then a hybrid that combines the strengths of both.

The exercise encourages it to compare different approaches instead of treating the first reasonable answer as the finish line. Better still, it gives you something far more valuable than a single recommendation. It gives you a range of possibilities and explains the tradeoffs between them.

Don't just skip to the finish line

The biggest advantage of this technique is that it makes the conversation about exploring alternatives rather than just picking a winner. Any keyword search can give immediate answers, but AI chatbots are more interesting when laying out competing ideas.

Imagine you are planning a weekend trip, often a go-to experiment. A standard prompt might produce a sensible itinerary filled with the highest-rated attractions. The three-solution prompt, meanwhile, begins with the expected museums and restaurants, then suggests renting bicycles to explore overlooked neighborhoods or planning the entire weekend around independent bookstores and local festivals. The hybrid version could blend a couple of famous attractions with enough unusual stops to make the trip feel personal.

The explanations are often as useful as the answer itself. Once you understand why ChatGPT prefers each option, you can make better decisions rather than simply accepting the recommendation with the nicest wording.

Hybrid conventionality

It's also a good prompt for iterating ideas. If the hybrid solution feels close but not quite right, you can ask ChatGPT to repeat the exercise using that version as the new starting point. After two or three rounds, the ideas often become noticeably more distinctive without drifting into complete nonsense.

The trick also scales well from short projects at home to more grandiose schemes that will take months to complete. Almost any situation that benefits from weighing different approaches can benefit from this style of prompt.

You can improve the results even further by giving ChatGPT a little context before asking for the three versions. A conventional answer built around your actual circumstances is much more useful than a generic one, and the unconventional suggestion becomes more interesting because it has meaningful boundaries to push against.

None of this guarantees a brilliant idea every time. Sometimes the unconventional option is genuinely impractical, and occasionally the hybrid answer feels like an awkward compromise. But the best prompts rarely force ChatGPT to mimic greater intelligence as much as encourage different approaches to problems. Asking for the conventional solution, the unconventional solution, and the best combination of both is a simple habit for more thoughtful conversations with AI chatbots.

DJI Osmo Pocket 4P: One-minute review

The Pocket 4P marks a big moment in DJI’s hugely popular series of compact vlogging cameras, being the first model with two lenses, even if Insta360 stole its thunder with the Luna Ultra, which it released in June.

I’ve already reviewed the single-lens Pocket 4, and it’s a brilliant handheld 4K camera. The ‘P’ version is largely identical but for its dual-lens array and a ‘refined’ 1-inch sensor that's equipped with LOFIC tech, plus a new D-Log 2 color profile.

@techradar

♬ original sound - TechRadar

For the most part, you can discover all you need to know about the Pocket 4P's design and performance in my Pocket 4 review. In this write-up, I'll focus on the new features of the bulkier model, and the real difference they make.

And I can say this right now: now that I’ve tried the Pocket 4P, it’ll be hard for me to return to single-lens models like the Pocket 4. That additional 3x telephoto lens unleashes a wide variety of shooting options to complement the highly capable ultra-wide main camera. On-the-go sequences with DJI's superb subject tracking in play, for example, can be particularly dynamic. It really is like having two cameras in one tiny device.

Two lenses means twice the camera (Image credit: Tim Coleman)

That being said, there is an inevitable design compromise: twin lenses make for a bulkier camera unit, with inexplicably no protective housing currently available, and a top-heavy design. If you already own a Pocket, and are wondering if the new Pocket 4P is worth it, or if it’s worth jumping ship from DJI to the Insta360 Luna Ultra, check out my Pocket 4 vs Luna Ultra article.

For me, the first-generation Luna Ultra is a compelling alternative with its 8K video and detachable module, but I can also tell that the Pocket 4P is a fourth-generation product — it feels more polished overall. It has neat accessories of its own, too, such as the magnetic fill light and remote module. As such, DJI's dual-lens Pocket 4P is now my top compact vlogging-camera pick.

DJI Osmo Pocket 4P: price and release date

• Announced on May 14 2026 at Cannes Film Festival, available from July 30
• Available in black or white versions, priced from £529 / AU$959. No US availability at launch
• Variety of bundles available, including a Vlog Combo kit for £605 / AU$1,099

With twin sensors / lenses, the Pocket 4P's mark-up over the single-lens Pocket 4 is justified. The 'P' version is excellent value. (Image credit: Tim Coleman)

Despite essentially having no competition in markets such as drones and, previously, gimbal vlogging cameras, DJI has always priced its products competitively. It usually undercuts rivals where actual alternatives exist, too, such as GoPro and Insta360 action cameras.

Insta360 outright said it wasn’t going to compete on price, but rather on features when speaking about its Luna Ultra, and that has turned out to be the case. Where the Luna Ultra starts at $770 / £650 / AU$1,230, the Pocket 4P comes in at £529 / AU$959 — though it's not available in the US at launch. Meanwhile, the single-lens Pocket 4 costs from £429 / AU$749.

Considering that the Pocket 4P has the extra camera with its own sensor, that price increase over the Pocket 4 feels completely justified, and in my book the Pocket 4P is excellent value, especially when you consider the quality of the 4K video footage it's able to produce.

The Pocket 4P is available in Black or White versions, and there are bundles available. The Vlog Combo, for example, includes the DJI Mic 2 Mini wireless mic, remote controller, fill light, tripod stand and tripod grip, and costs £605 / AU$1,099 — that's the bundle I'd go for.

• Price score: 5/5

DJI Osmo Pocket 4P: specsDJI Osmo Pocket 4P specs

Lens(es):

20mm f/2.0 (1-inch sensor), 60mm f/1.8 (1/1.28-inch sensor)

Video:

4K and 1080p up to 60fps, plus slow motion up to 240fps

Photo:

Up to 9.4MP in RAW & JPEG, up to 37MP in SuperPhoto mode

Storage:

107GB internal, microSD card slot

Battery:

1,545mAh, up to 3.5 hours Full HD record time

Charger type:

USB-C / optional fast PD charger

Weight:

8.1oz / 230g

Dimensions:

159.5 x 63.3 x 33.5mm (L x W x H)

DJI Osmo Pocket 4P: Design

• Dual 20mm f/2.0 and 60mm f/1.8 lenses
• Otherwise the same design as the Pocket 4, with rotating screen and gimbal-mounted camera
• USB-C charging and 107GB of internal storage in addition to micro SD

(Image credit: Tim Coleman)

Save for the bulkier dual-camera system and a total weight of 8.1oz / 230g, the Pocket 4P’s design is identical to the Pocket 4; it’s extremely compact, features a 2-inch touchscreen which rotates 90 degrees between vertical and horizontal formats, and a customizable five-button array, with a joystick that controls gimbal movement or zoom.

The extra two buttons, revealed when the screen is rotated to horizontal, help speed up access to various settings. For example, the C button can be assigned for direct access to two custom profiles — handy if you regularly switch between, say, color settings such as D-Log and shooting modes such as slow motion.

It certainly pays dividends to spend time customizing the camera to suit how you use it, so that you’re not fussing with settings when you're out and about.

For me, the single-lens Pocket 4 feels perfectly balanced, and it’s notably smaller than Insta360’s Luna cameras. The 4P, however, somewhat sacrifices that advantage on the alter of dual-camera versatility — it’s a little top-heavy, and I found it more comfortable to use with a tripod grip and mini tripod attached. The Luna Ultra feels more balanced because its handle is bulkier (see the Pocket 4 / 4P / Luna Ultra comparison photos, below).

Gimbal movement still feels smooth, and the shots steady and smooth, despite the extra weight of the cameras.

DJI Pocket 4 (left), DJI Pocket 4P (right)Tim ColemanDJI Pocket 4 (left), DJI Pocket 4P (right)Tim ColemanDJI Pocket 4P (left), Insta360 Luna Ultra (right)Tim ColemanDJI Pocket 4P (left), Insta360 Luna Ultra (right)Tim ColemanDJI Pocket 4P (left), Insta360 Luna Ultra (right)Tim Coleman

At launch there's no hard-shell protective case for the Pocket 4P, as there is for the Pocket 3, nor a slimline gimbal protector, like you get with the Pocket 4. This means the bulkier camera module remains somewhat unprotected, which is something DJI needs to be rectify ASAP.

There are also no dedicated ND filter kits from DJI at launch, though I fully expect that situation to change very soon through third parties (it might even have done so by the time you read this).

As with the Pocket 4, there’s the fill-light accessory, which attaches magnetically to the gimbal’s arm and is powered by the camera, and which lets you choose between three brightness levels and three color temperatures.

The light won’t make much of a difference to your images in bright daylight, but in low light and for indoors, it’s game changing for vlogging. I found the warm setting at the medium brightness level was a sweet spot for the times I wanted to use the light.

The Pocket 4P turned off, where its lenses sit out horizontallyTim ColemanAnd turned on, the lenses sit verticallyTim ColemanThe main 1-inch ultra-wide 20mm f/2.0 camera (bottom), and the new 3x telephoto f/1.8 camera (top)Tim ColemanThe screen rotates 90-degrees between vertical and horizontal formats. When shooting vertically, two control buttons are hidden by the screenTim ColemanBut they are revealed in horiztonal formatTim ColemanNote the top gimbal arm is magnetic and accepts a fill light accessory, which is powered by the camera and has three brightness settings, and three color temperature settingsTim Coleman

Then there’s the tiny remote control, borrowed from the Osmo Mobile 8P. This easily connects to the Pocket 4P over Bluetooth, and can be used to remotely view and control the camera. It’s a neat answer to Insta360’s modular controller in its Luna cameras, though the DJI module doesn’t feature a built-in mic, as Insta360’s does.

For occasions when you're taking hands-free shots with the Pocket 4P rested on a surface or support, the remote controller is super-handy. The controls are slightly stripped back versus the camera’s control layout, but the key controls, such as touch subject-tracking, record, and gimbal movement / zoom are there.

• Design score: 4.5/5

DJI Osmo Pocket 4: Performance

• Improved subject tracking
• 107GB of built-in memory added
• Higher-capacity battery extends record times

A major feature in Pocket cameras is subject-tracking autofocus. You really don’t want to be messing around moving the gimbal manually with the joystick, which can be fiddly and jerky. Thankfully, subject tracking, the latest version of which DJI calls ‘ActiveTrack 7.0’, is super-reliable — you double-tap on the subject, and the camera does the rest. For me, it performs slightly better than Insta360 Luna’s tracking, even if it does at times feel like its ‘keeping up’ with fast-moving subjects rather than keeping them central.

Battery life is solid, though you will notice it drop quicker with certain modes, such as 4K 60fps video recording, hyperlapse, and slow-motion 240fps. The camera also gets pretty warm pretty quickly when using these modes, though I’ve not had it overheat and shut down during my testing.

Future / Tim ColemanTim Coleman

With compact vlogging cameras such as the Pocket series, the battery is non-removable, so you are relying on that USB-C rechargeable internal unit. I found myself conscious of having to charge the camera on heavy use days — that’s life with a Pocket. There is, however, an optional battery grip available to extend shot life.

In addition to a microSD card slot, DJI includes a generous 107GB of built-in storage. That’s enough for a lot of 4K clips, even when utilizing the maximum 200Mbps bit rate.

Recording internally, you'll need to transfer files either over Wi-Fi or wired USB-C. Thankfully, there’s next-gen Wi-Fi 6 and USB 3.1 support on board, delivering fast transfer speeds and making for an efficient workflow on the move.

As with other Pockets, the 4P can get a little warm when you're using its more power-hungry features, such as 4K 60fps and hyperlapse. I've not had any overheating issues to the point where the camera shuts down, but I was aware of the camera warming up while holding it.

• Performance score: 5/5

DJI Osmo Pocket 4: Image and video quality

• 1-inch sensor ultra-wide camera, and a new 3x telephoto, up to 12x zoom
• D-Log2 profile with up 17EV dynamic range
• Low-light further boosts the improved 14 stops dynamic range

There is one minor difference between the Pocket 4P and Pocket 4’s ultra-wide 20mm f/2.0 camera: the ‘refined’ 1-inch sensor, which unleashes a new log color profile, D-Log 2.

DJI says the refined sensor features LOFIC tech — which in simple terms is like super HDR, only it works in real time, which differs to HDR. I first came across the term LOFIC when reviewing the Oppo Find X9 Ultra phone.

The fruit of LOFIC and D-Log 2 is an expanded 17 stops of dynamic range, while the regular Log profile has around 14 stops. D-Log 2 is only available with the ultra-wide camera because it’s linked to the sensor tech, whereas the new 3x telephoto camera uses a separate 1/1.28-inch sensor.

D-Log 2 also isn’t available in every shooting mode, such as Slow Motion 240fps, although like the Pocket 4, that 10x slo-mo is available in 4K, or up to 200fps when using the 3x telephoto camera.

I’ve shot the same clips in D-Log 2, D-Log and ‘Normal’ color profiles (all 10-bit) to make quality comparisons, especially regarding dynamic range, and to see how easy it is to grade the footage for a consistent look. I’ve also shot using some of the built-in color profiles, such as ‘NC film’ and ‘Movie’.

The level of detail that the Pocket 4P is able to hold, especially in D-Log 2, is majorly impressive, especially for such a small and affordable camera. For me, outright 4K image quality is where DJI has the edge over Insta360, and most leading smartphones.

Is D-Log2 enough reason to upgrade from the Pocket 4? Not for me, but there is a more compelling reason: the second camera. The 60mm f/1.8 telephoto optic is essentially a 3x portrait lens, with a pleasant shallow depth of field for upper-body vlogging and shooting closeup details. It’s limited to D-Log or Normal color profiles, but the overall quality is really, really nice, and the footage gives a totally different perspective to the ultra-wide camera's.

Where that 3x lens really shines is when it’s coupled with some of the Pocket 4P’s major features, such as active tracking. With the tighter perspective, creative camera movements tracking a subject on the move can look particularly dynamic, especially where there are other objects in the foreground.

The zoom capabilities are further extended — where the Pocket 4 was limited to a 2x lossless zoom and 4x digital zoom, the Pocket 4P again has the 2x lossless zoom with the ultra-wide lens, plus 6x lossless with the telephoto lens and up to a 12x digital zoom.

The quality of that 12x zoom is okay at a pinch, but if quality is your main concern, I wouldn’t go beyond the 6x setting.

I don’t like the way the heavier dual-camera setup impacts how the camera feels in the hand, but it’s a game-changing feature for this type of camera, and for me it now feels impossible to return to the single-lens format.

The Luna Ultra goes one step further for resolution, with 8K video recording. Personally, I’m happy with a camera that is limited to 4K recording, but whose quality at this resolution, especially regarding dynamic range and color, is more pleasing on the eye. That's the Pocket 4P.

• Image and video quality score: 5/5

DJI Osmo Pocket 4P: testing scorecardDJI Osmo Pocket 4P

Attributes

Notes

Rating

Price

Pricier than the Pocket 4 but cheaper than the Luna Ultra. You're getting a lot of camera for your money

5/5

Design

The dual lenses are a success, but in otherwise keeping the same body design, the Pocket 4P feels top-heavy

4.5/5

Performance

Industry-leading image stabilization and subject tracking, decent battery life and generous internal storage

5/5

Image and video quality

Yes it's 'only' 4K, but the Pocket 4P can shoot gorgeous footage up to that resolution, especially with the second lens

5/5

Should I buy the DJI Osmo Pocket 4P?Buy it if...

You want more from your Pocket
The dual lenses are a revelation, broadening the kind of shots you can make versus the single-lens models that are restricted to an ultra-wide field of view.

You want the most polished and feature-rich compact vlogging camera
Sure, the Luna Ultra comes with innovative features, but the Pocket 4P has its own unique tricks, and overall it feels like the more polished camera.

Don't buy it if...

You don't need the second lens
Many people have happily been vlogging with a single-lens Pocket camera for years. If you only need that ultra-wide angle, it's wiser to stick with the Pocket 3 or opt for the Pocket 4 instead — they're physically smaller, and cost less.

You want the smallest possible Pocket
The dual-camera setup wins for versatility, but it does result in a bulkier, top-heavy design when compared to the regular Pocket 4.

DJI Osmo Pocket 4P: also consider

DJI Osmo Pocket 4

Don't need the second lens? You'll save a packet by opting for the Pocket 4 instead, which otherwise has essentially the same features, and get a more streamlined camera, with a neat gimbal protector to boot.

Read our in-depth DJI Osmo Pocket 4 review

Insta360 Luna Ultra

Love the sound of dual lenses in a compact vlogging camera? The Luna Ultra is a compelling and strikingly similar alternative, but with 8K video recording and a neat detachable module which comprises the screen, key control and built-in mic. The Luna Ultra arguably wins on innovation, but the Pocket 4P wins by being more polished. The Luna Ultra is pricier than the 4P, too.

Read our in-depth Insta360 Luna Ultra review

How I tested the DJI Osmo Pocket 4P

(Image credit: Tim Coleman)

• DJI loaned me the Pocket 4P Vlog Combo kit for a couple of months ahead of the July 30 release
• I shot plenty of 4K footage with the various color profiles, and used the various accessories
• I compared it to the single-lens Pocket 4 and dual-lens Insta360 Luna Ultra

I used the DJI Pocket 4P for almost two months ahead of my review embargo, during which time I regularly shot 4K video using the 20mm and 60mm lenses, the various frame rates up to 240fps, and the range of color profiles, including the new D-Log2. I tried the various shooting modes too, such as hyperlapse, and took photos in the regular and SuperPhoto modes.

In the Vlog Combo kit was a DJI Mini 2 Mini, remote module and fill light, and I tested both those accessories. I generally recorded onto the camera's internal memory, transferred files through USB-C and checked transfer times, ran the camera battery down, and checked recharge times using the supplied USB-C cable.

During my review period I also had the Pocket 4, enabling me to assess the key differences between the single- and dual-lens models. I also had the rival Insta360 Luna Ultra, again allowing me to make direct comparisons.

• First reviewed: July 2026
• Read more about how we test

The problem with online shopping is that it makes it far too easy to buy pretty much anything, without properly considering whether you really need it. I started to wonder if ChatGPT could help me make better decisions, by questioning every purchase I was going to make, before I made it.

To test ChatGPT’s ability to stop me wasting money, I picked four tempting purchases and asked ChatGPT to argue against each one. It had to check the price history, suggest cheaper alternatives, and decide whether the purchase solved a real problem, or if I was just scratching an itch to buy.

Here's how ChatGPT did — and the results surprised me.

GPT before you buyDid you know TechRadar now has membership?

(Image credit: Future)

Become a TechRadar Insider by simply clicking 'Join Now' at the top of this page. Have a question? Please email membership@techradar.com

I started with a tech gadget I was thinking of buying — an Apple MacBook M5 — and went to ChatGPT with the following prompt: “I'm thinking of buying an Apple MacBook Air 13-inch Laptop M5 chip. Help me check the price history, suggest cheaper alternatives and decide whether the purchase solves a real problem."

ChatGPT said it would check current pricing and recent lows, compare genuinely cheaper options, then pressure-test whether the purchase replaced a real limitation, or was mostly me trying to satisfy an upgrade itch, and off it went.

After a lot of thinking, ChatGPT replied with a devastating verdict that made me pull back from hitting the 'Buy' button: “Don’t buy the M5 MacBook Air yet.”

It suggested I get the previous M4 version instead, which is considerably cheaper, which didn't hugely surprise me. Then it gave me some questions to ask myself, in order to work out if buying a new MacBook was actually solving a real problem, or if I was just satisfying a buying itch.

Questions like, “Does your existing Mac have noticeable slowdowns?” It suggested I shouldn’t buy a new machine if my main argument was simply that my current Mac was a few generations old.bChatGPT also gave me some buying options if I was determined to go through with a purchase.

The most useful advice here wasn’t the cheaper recommendation — it was being forced to identify the exact limitation my current MacBook was causing me. I couldn’t, so I decided to keep it until I could.

Then I hit on a simple phrase that could potentially work even better — "Try to talk me out of it".

Talk me out of it

I repeated the experiment with a few other items I’d been thinking of buying recently. An item of clothing (new trainers), a subscription (Disney+), and a kitchen thing (a new microwave). But this time I added "Try to talk me out of it" at the end of each prompt.

In each case, Chat surprised me with its answer, provided alternatives, and made me question whether I genuinely wanted the item, or if there was something else driving my decision.

With the trainers, it pointed out that I already owned shoes that served the same purpose, and suggested waiting until those wore out. For Disney+, it recommended subscribing for a single month when there were several things I actually wanted to watch. The microwave was different — because our existing one had a genuine fault, ChatGPT concluded that replacing it was justified.

The experiment worked because it helped me mentally reframe each purchase. Instead of asking whether I wanted something, I had to consider what problem it solved, what I already owned, and whether there was a cheaper way to get the same result.

Online shopping is designed to remove as much friction as possible from spending money. ChatGPT gave me some of that friction back. So now, whenever I’m about to spend a few hundred pounds, or a few thousand, I ask it the same question: “Talk me out of it.”

Artificial intelligence came at just about the right time, speeding up app and software development as the world started to contend with skills shortages, but it changed the pace so much that security teams have not been able to keep up.

Recently, we’ve seen AI being applied across multiple other domains with role-specific agents and tools, but that’s introduced its own challenges. While tools like Claude Code have proven a hit for generating, reviewing and editing code in seconds, security-focused tools like Anthropic’s Claude Mythos family of models are having broader impacts on the industry.

Anthropic itself has even admitted that Mythos is so powerful that the worry it could be abused by malicious criminals is extremely real – the Preview model is currently only available to a select number of pre-approved partners.

So with AI now capable of inspecting code, discovering vulnerabilities and suggesting fixes, do organizations even need as many human workers on the case, or can they get by with significantly fewer humans in the loop serving as AI reviewers? Recent layoffs have certainly implied as much.

The evolving role of security workers in an AI-first world

But with the entire lifecycle of development now amplified by AI, experts are warning that companies could actually be creating more work for themselves, and more than they could ever handle, leaving them facing strains from angles they weren’t previously exposed to.

For example, fewer than one in 10 companies now fix 90% of identified vulnerabilities within 90 days – implying that the volume of vulnerabilities is indeed increasing, rather than that fix efficiency is slipping.

Anthropic even revealed that its around 50 early Mythos Preview partners discovered more than 10,000 high- or critical-severity vulnerabilities – and thousands more of lesser significance.

Checkmarx CEO Sandeep Johri predicts we could soon find a balance, where vulnerabilities volume matters less and we revert our focus back toward exploitable risks. I spoke with Johri about the evolution of AppSec, where AI is and isn’t useful, and how organizations can balance speed and control.

• With the rise of AI coding tools and AI-generated software, some are questioning whether traditional application security practices are becoming outdated. Is AppSec actually becoming obsolete, or is it evolving?

Traditional application security is not obsolete. It is evolving to meet the reality of how software is being built today.

For years, the process was fairly linear: developers wrote code, security teams scanned it, and vulnerabilities were addressed later. That approach becomes much harder when software is being created at a much faster pace with the help of AI.

AI accelerates development and risk simultaneously: 70% of developers say AI-generated code created more vulnerabilities in 2025, according to our research. As code volume and complexity compound, security needs to move earlier into the development process, giving developers the tools and guidance they need while they are building.

Security teams will continue to play a critical role to help organizations develop software and maintain confidence in their enterprise applications. But their focus needs to shift from finding vulnerabilities to remediating them at scale, because we are tracking an enormous gap in most companies. Our data finds that fewer than 10% of organizations fix 90% of identified vulnerabilities in 90 days.

• AI coding tools are helping developers create software faster than ever before. What new security challenges does this introduce for organizations adopting these technologies at scale?

The biggest challenge is that development speed is increasing faster than many security processes can keep up with. AI coding tools allow teams to create and deploy software quickly, but the code generated by AI still needs to be reviewed, tested, and secured.

Companies that ship 81-100% of their code with AI are nearly three times more likely to ship vulnerable code than those who use AI 1-20% of the time. This volume can overwhelm security teams with thousands of findings, many of which don't represent meaningful risk. The priority needs to be identifying the vulnerabilities that actually create exposure and helping teams fix those issues faster.


• Many organizations are looking to AI to help identify and fix security vulnerabilities. Why shouldn’t companies rely solely on AI models to secure the code that AI is helping create?

AI is a valuable tool for security teams, but organizations still need accuracy, context, and human oversight. AI can help identify patterns, analyze code, and accelerate remediation, but security decisions require confidence in what risks actually matter.

Frontier models can uncover hidden exploit paths, but they can also deliver inconsistent findings and false positives. Their results may change depending on the prompt, and they can still miss known critical vulnerabilities.

The challenge with relying only on AI is that organizations may create a false sense of security, or “automation bias.” AI models can generate code and help analyze vulnerabilities, but they need to be paired with security expertise and proven security practices.

The most effective approach combines AI-driven capabilities with strong security foundations, so teams can move faster while maintaining control over risk.

• As companies adopt more AI tools throughout the development process, what are the biggest security risks they need to consider beyond just AI-generated code?

Organizations need to think beyond the code itself and look at the entire AI ecosystem being introduced into software development. Many companies are adopting AI tools, models, agents, libraries, and other components faster than they can establish governance around them. This creates visibility challenges because security teams may not know what AI technologies are being used, where they exist in applications, or whether they meet security requirements.

Another concern is shadow AI, where employees use AI tools without formal approval or oversight. Organizations need visibility, clear policies, and a way to manage these technologies as part of their overall software supply chain.



Perhaps the most urgent problem is the expansion of the attack surface itself. With LLMs, it has never been faster, cheaper, or easier for bad actors to exploit software. Issues that sat undetected for years are now being surfaced and weaponized at machine speed. Of the vulnerabilities Mythos has found so far, 99% haven't been patched, according to Gartner.

• How does the rise of AI change the role of security teams? Does the traditional approach to finding vulnerabilities need to shift toward a model focused more on prioritization, remediation, and continuous protection?

Identifying vulnerabilities is no longer enough when organizations already have more findings than they can realistically address. Security has to become continuous, embedded in development workflows, working in lockstep with developers, to build securely from the start while maintaining visibility and control.



We are shifting the focus to understand which issues create the greatest risk to give developers the context to address them fast, where code is written in the IDE.

Fidelity now matters more than volume. One verified true positive is worth more than a hundred low-confidence findings. If developers can’t trust what they’re shown, they’ll start ignoring it. That’s why organizations are increasingly looking at metrics like F1 score, which measure precision and recall together, rather than raw finding counts.

• What does the future of application security look like in an AI-driven software development world? Will organizations need a different approach to balancing speed, innovation, and security?

The future of application security will require a more integrated approach. Organizations are going to continue adopting AI because the productivity benefits are significant, but security needs to evolve alongside that innovation.

Security will become more agentic, more intelligent, and more closely connected to the development process. Part of that evolution is combining deterministic, rules-based scanning with AI-driven reasoning in a single process, rather than running them as separate, disconnected tools. Deterministic methods catch what’s already proven; AI reasoning catches what’s novel. Together they’re more complete than either alone.



In addition, deterministic models have real cost advantages. Asking a frontier model to reason its way to security (i.e. extra review passes, self-generated threat models) burns tokens fast. That cost compounds the longer a vulnerability survives: cheap to fix in the IDE, more expensive in CI/CD, most expensive once it's live in runtime. And every time a developer has to stop and pull a vulnerability out of code that's already shipped, that's velocity lost to rework instead of innovation.

Teams will need technology that can help identify real risks, support faster remediation, and provide visibility across the entire software lifecycle. The organizations that succeed will be those that make security part of how they build software, allowing developers to move quickly while reducing unnecessary risk.

• For organizations that are embracing AI coding tools today, what steps should they take to make sure they can innovate quickly without introducing unnecessary security risks?

The first step is visibility. Organizations need to understand where AI is being used, what tools are being introduced, and what impact those tools have on their applications.

Remediation is far cheaper the earlier it happens — catching an issue in the IDE costs a fraction of catching it further down the pipeline. But there’s another unsettling gap in our research: nearly all developers have access to in-IDE security tools, but fewer than one in five actually secure code as they write it. The cost of fixing that issue compounds as it passes through later stages of development.



In addition, organizations need clear governance around AI adoption, because only 22% currently have formal AI governance policies in place. That means defining policies, monitoring usage, and making sure teams have the right security controls as they continue to innovate.



AI will continue to change software development. The companies that benefit most will be the ones that embrace the technology while building security into the process from the beginning.