Feed aggregator

European Commission President Ursula von der Leyen on Wednesday proposed making Canada the European Union's first associate member, a striking overture as U.S. President Donald Trump's tariffs drive Ottawa closer to Europe.

(Image credit: Pascal Bastien)

In July 2026, an OpenAI model being tested in a security research environment broke out of its sandbox, exploited a zero-day vulnerability and autonomously compromised systems at Hugging Face, executing more than 17,000 attacker actions in under five days with no human directing the attack.

It’s the most autonomous, most damaging agentic AI attack documented to date, and it signals a significant shift from AI-assisted hacking to fully autonomous cyber operations.

It isn’t the first: Anthropic disrupted a similar Claude Code-driven espionage campaign, which it attributed with high confidence to a Chinese state-sponsored group, some ten months earlier, and Sysdig documented the first fully agentic ransomware attack just over a week before the Hugging Face intrusion began.

But it’s the starkest proof yet that autonomous AI attacks have moved from lab hypothesis to live threat. This event was a wake-up call not just for the security industry but for everyone with an online presence, and the implications are sobering.

Up until this point, discussions around AI-powered cyber threats had largely focused on how AI can help attackers work faster and at scale. As an industry, we discussed how the technology could write more convincing phishing emails, analyze larger datasets, or accelerate malware development. But the OpenAI case points to something far more troubling: AI carrying out the attack itself, from start to finish.

The old cybersecurity playbook no longer works

The OpenAI incident shows how quickly the threat landscape is evolving, and how exceptionally adept agentic AI now is at finding and exposing existing flaws. It’s a stark illustration that the very tools helping to fight attacks can also cause them. The situation shows how vulnerabilities that once may have slipped through the cracks are now becoming far easier to find and exploit.

That’s a problem for the sector as a whole. But it’s an even bigger problem for organizations that continue to overlook the basics of cyber hygiene. Despite the security industry’s best efforts, known weaknesses, unpatched systems and outdated software remain the norm rather than the exception.

Verizon’s 2026 Data Breach Investigations Report found that vulnerability exploitation overtook stolen credentials as the leading breach vector for the first time in nineteen years, accounting for 31% of breaches, while organizations patched only 26% of flaws on CISA’s Known Exploited Vulnerabilities list, down from 38% the year before. Too many organizations still treat security tools as a last line of defense rather than closing the gaps before attackers, human or artificial, get there first.

If Hugging Face’s encounter with a rogue AI agent has taught us anything, it’s that a reactive security posture is no longer sustainable. As AI becomes more capable of identifying and exploiting weaknesses at machine speed, IT leaders have far less time to detect and respond before a minor flaw becomes a major breach.

The UK’s NCSC and its Five Eyes partners said as much in their May 2026 joint guidance on agentic AI: the question every organization must now answer is whether it can understand, monitor and contain what its AI agents, and the ones attacking it, actually do.

In practice, that means shifting focus from detecting attacks to reducing the opportunities for them to succeed in the first place, building continuous visibility, effective governance and the ability to act before minor vulnerabilities become major attack paths, rather than relying on periodic checks and reactive response.

The need for real-time visibility

After all, you cannot defend what you cannot see. As IT environments become more complex, maintaining an accurate understanding of assets, vulnerabilities and configurations becomes increasingly difficult.

OpenAI’s agentic attack reinforces why this matters. If autonomous AI can identify and exploit vulnerabilities without human input and at machine speed, organizations can no longer rely on scheduled scans to understand their exposure.

Instead, real-time visibility provides a continuously updated picture of an organization's environment, allowing them to identify risks as they emerge rather than discovering them by chance and when it’s too late.

But updating technology is only half of the equation. Organizations also need strong governance so that they can understand who owns risk, how vulnerabilities are prioritized and how quickly issues are addressed.

And if attacks are being carried out by agentic AI, defenders must use AI to fight AI. Hugging Face’s own investigation shows why. When its frontier-model tools refused, on safety grounds, to help analyze the malware, the team turned to GLM-5.2, a Chinese open-weight model, running on its own infrastructure, and got the job done in hours instead of days.

That’s an uncomfortable irony for the Western AI industry, and a preview of the governance decisions security leaders will increasingly have to make. That’s why autonomous IT operations are increasingly taking center stage. Rather than replacing people, they enable teams to focus on strategic decisions while routine remediation, patching and risk reduction activities happen at the speed required to keep pace with modern threats.

Preparing for a new threat landscape

Whether OpenAI’s rogue agent turns out to be a one-off or the new normal, organizations can’t afford to wait and see. The threat landscape is evolving at machine speed, and defenders need to be ready for what’s coming next.

But organizations are not powerless. The technologies needed to improve resilience already exist; it’s now a question of adopting the right tools and using them effectively.

Going forward, businesses can no longer take a reactive approach to cybersecurity. The ones still on the back foot when the next agentic AI attack lands will be the ones counting the cost.

The OpenAI incident shows that the age of AI-assisted attacks is rapidly giving way to something more sophisticated. The direction of travel is clear, and the days of relying on security tools as a last line of defense are over. It’s time to get on the front foot and close the gaps before attackers, human or artificial, find them first.

We've featured the best secure smartphone.

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

At last, Joe (Zoe Saldaña) has been rescued in Lioness season 3. Frankly, that's an episode earlier than expected.

Still, it seems as though there's more to be solved. The timing of her rescue suggests there's more to the CIA's strategy than we've been able to give proper attention to (at least, I really hope that's the case), with Joe saying she can potentially "never" return home.

But does anybody really know what they're talking about? And when does Lioness season 3 episode 8 arrive on Paramount+?

What time can I watch Lioness season 3 episode 8 on Paramount+?

Lioness season 3 episode 8 will drop on one of the world's best streaming services in the US and Canada on Sunday, September 20 at 12am PT / 3am ET.

Here's when it will be released in other nations:

• US – 12am PT / 3am ET
• Canada – 12am PT / 3am ET
• UK – 8am BST
• India – 12:30pm IST
• Singapore – 3pm SGT
• Australia – 5pm AEST
• New Zealand – 7pm NZST

When do new episodes of Lioness season 3 come out?

(Image credit: Paramount+)

Lioness season 3 will have a total of eight episodes, with new entries airing weekly. That gives us the following schedule:

• Episode 1: out now
• Episode 2: out now
• Episode 3: out now
• Episode 4: out now
• Episode 5: out now
• Episode 6: out now
• Episode 7: out now
• Episode 8: September 20

For the past few years, the spotlight has been firmly on AI models.

Organizations have poured investment into generative AI, machine learning platforms and large language models. New capabilities appear almost weekly. Yet despite all this progress, a familiar question keeps surfacing in boardrooms and leadership meetings – where is the business value?

Part of the answer is that the technology itself is no longer the main obstacle. Powerful AI tools are now within reach of most organizations. What remains difficult is turning AI-generated insights into decisions and measurable business outcomes.

This is the last mile of AI, where many projects lose momentum. A model may produce a recommendation in seconds, but acting on it is often far more complicated. Data may be incomplete, critical context may sit elsewhere, and governance teams may not be confident in the output.

As a result, many organizations find themselves investing heavily in AI while struggling to move beyond pilots and proofs of concept. The missing piece is often the ability to connect intelligence to the reality of how the business really operates.

The real AI bottleneck

When executives talk about successful AI programs, they rarely focus on the sophistication of the model. What matters is the outcome. For example, whether fraud losses fall, if customers can be onboarded faster, or if downtime is reduced. Those outcomes depend on much more than AI itself.

For AI to create value, it needs a clear view of what is happening across the business, enough context to understand what those events mean, and the ability to operate within governance guardrails. When any of those elements are missing, recommendations become harder to trust, automation stalls, and adoption suffers.

Many organizations discover that the real bottleneck sits between the model and the business process. Data is fragmented across systems, and essential context is missing. Often, information arrives too late to be useful. Different teams define the same business concepts in different ways. These are not new problems, but AI has exposed them more clearly than ever before.

We see this challenge regularly in financial services. One institution had invested heavily in AI-driven fraud detection, yet investigators still struggled to act quickly because customer records, transaction histories, and external fraud signals were spread across multiple systems. Once those sources were brought together in a trusted, governed view, fraud losses fell, false positives declined, and onboarding processes became more efficient.

AI needs a broader view of the business

One reason many organizations struggle to get AI into production is that the technology depends on a much broader view of the business than traditional analytics ever did.

Most companies have spent years investing in ERP platforms, CRM applications, and operational databases. These systems remain the authoritative record of the business and provide much of the information AI needs to work effectively.

To make good decisions, AI increasingly needs information from outside those core systems. It may need supplier data to understand a disruption in the supply chain. It may need information from partners, SaaS applications, or external intelligence sources to provide business context. Increasingly, it also needs access to live operational signals, whether that's customer interactions, fraud alerts, connected devices, or event streams that reveal what's happening right now.

You can think of this as three layers of information:

• The first is the authoritative data held within core enterprise systems.
• The second is contextual information from partners, suppliers and external sources that helps explain why events are occurring.
• The third is real-time operational awareness, the signals that show what is happening in the moment.

Most organizations have made significant progress managing the first layer. Far fewer have found an effective way to combine all three. A global manufacturer we worked with faced a similar challenge.

Its AI models could predict equipment failures, but critical information about production schedules, supplier delays, and maintenance activities sat across different systems. Connecting those sources gave teams the context needed to identify risks earlier, reduce downtime, and make better operational decisions.

As a result, AI often operates with an incomplete picture of reality. It may understand what happened yesterday, but not what is happening now. It may have access to internal records, but lack the external context needed to make a confident recommendation.

This is where the last mile challenge starts to emerge.

The ecosystem effect

The rise of AI is also changing how organizations think about data sharing. For years, most businesses focused primarily on their own internal systems. Today, many of the decisions that matter depend on information that sits beyond organizational boundaries.

A manufacturer responding to supply chain disruption needs visibility into suppliers and logistics partners. A bank trying to detect fraud benefits from external signals as much as internal customer data. Public sector organizations often need information from multiple agencies to improve services and outcomes.

The common thread is that intelligence increasingly flows across an ecosystem rather than a single enterprise. Trusted AI outcomes depend on trusted access to data wherever that data resides.

Technology leaders therefore face a balancing act: making data accessible across distributed environments while maintaining governance, security and ownership. Without that foundation, even the most advanced models struggle to deliver consistent results.

Closing the last mile

If there is one lesson emerging from the first wave of enterprise AI adoption, it is that success depends less on deploying another model and more on creating the conditions that allow AI to operate effectively.

That starts with trusted data. Business users need confidence that the information feeding AI systems is accurate, up to date, and governed consistently. The challenge is also making it available in a way that AI can use without creating new silos or adding more complexity. AI needs access to core business systems, external ecosystem data, and real-time operational signals. If those remain disconnected, the quality of the output will always be limited.

Why understanding matters

Data alone, however, rarely tells the full story. AI needs to not only understand the data itself, but the business context around it. For example, where it came from, how it relates to other information, and the policies that govern its use. That context is often what separates a useful recommendation from a misleading one. Without it, even accurate data can lead to poor decisions.

This is where active context comes in. It provides AI with the richer understanding needed to interpret information correctly, connect it to the wider business environment, and make decisions that are grounded, relevant, and well governed.

Turning insight into action

The final step is where many organizations still struggle. Insights need to become actions.

Too often, AI generates recommendations that sit in dashboards, reports or isolated applications. The greatest value comes when intelligence is embedded directly into operational workflows, helping people make decisions faster or automating routine processes altogether. That is where AI starts to move beyond experimentation and deliver measurable business outcomes.

For technology leaders, a useful test is to ask a handful of simple questions.

• Can AI access information as events happen, or is it relying on yesterday's data?
• Can it combine internal and external sources to build a complete picture?
• Are governance policies applied consistently across different environments?
• Do people trust the outputs enough to act on them?

If the answer to any of those questions is no, the next investment should probably be focused on the data foundation rather than the model itself.

Where the next gains will come from

The organizations generating the greatest returns from AI are not necessarily those deploying the newest technology. More often, they are the ones that have found a way to connect trusted data, business context, and operational processes.

In banking, that might mean reducing fraud losses while accelerating onboarding. In manufacturing, it could mean identifying disruptions earlier and reducing downtime. In the public sector, it may involve improving citizen services through better collaboration across departments. In each case, the value comes not from the model itself, but from embedding intelligence into operational decisions and workflows.

That is the last mile of AI. For many organizations, it remains the biggest barrier to success. It may also be the biggest opportunity.

We've featured the best AI chatbot for business.

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

On our podcast I’m known as the Apple hater, but over the past year I’ve let up a little — even defended the company.

I’ve said Apple could be the only company capable of releasing camera glasses without major backlash thanks to privacy-first approach, celebrated its approach to AI as possibly genius if the bubble does pop, and most recently said the iPhone Duo looks great and that the visible crease in some shots should be ignored — as if it’s like the Z Fold 8 Ultra you’ll never see it while you’re using it.

But in our most recent episode, I saw something that I couldn’t believe in the new AirPods 5 specs. I had to actually message our Audio Editor because I thought I must be wrong.

How are on-bud volume controls only now coming to AirPods? And how are they reserved for the pricier model?

Take back (volume) control

I brought this up in the podcast, but my Apple-wielding colleagues — Axel Metz and Matt Evans — argued the feature isn’t that useful anyway. Those with Pro-level AirPods that boast volume controls noted that they rarely use them

Folks, that’s beside the point.

Volume controls are one of the most basic earbuds functions. Take a minute to scroll through Amazon, and you can find dozens of wireless earbuds under $20 that offer volume controls. This shouldn’t be something reserved for the fifth generation of a premium earbuds brand like AirPods. It’s absurd!

Oh, and I think volume controls on the buds are useful, for what it’s worth. If I’ve set my phone up to watch a YouTube video and the volume spikes, I want to be able to quickly swipe down without reaching over; or I might want to change the volume of my workout tunes mid-session without pulling my phone out and risking getting distracted.

Sure, it’s not a game changer, but as I said already, it’s a basic feature we should expect. It would be like if your iPhone didn’t have an email app — there are ways around it, but why cut off the root of the problem?

Perhaps I shouldn't give Apple ideas (Image credit: Jacob Krol/Future)

What really grinds my gears, however, is that volume controls still aren’t truly standard. If you get the base $129 / £119 / AU$219 AirPods, you’ll be out of luck — you need to pay extra for the $149 / £139 / AU$249 model with a wireless charging case.

This isn’t anywhere near as egregious as those ridiculously expensive Mac Pro wheels from early 2020 — back when we could worry about such trivialities in the ‘before times’ — but it feels like a remnant of Apple’s arrogance, and the quality that has put me off the company’s tech for all these years.

People can get the earbuds they want, and I’m sure the AirPods 5 will be just as solid as the previous generation. But when there are so many excellent audio options out there, my philosophy is to get a pair that respects you, offers genuine value for money, and won’t nickel-and-dime you for features that should have been added back in the first-generation model — not as an upcharge on the fifth-generation iteration.

• Elastic Security Labs uncovered REF9334, a Brazilian banking malware campaign active since May 2025
• Malware “Kremlin” deploys fake docs and malicious Chrome/Edge extensions to steal banking data
• 1,515 infections found, 98% in Brazil

Security researchers from Elastic Security Labs have discovered a new Brazilian banking malware campaign that uses browser extensions to compromise users and steal sensitive information.

In an in-depth report published earlier this week, the researchers said the campaign has been active since at least May 2025. Dubbed REF9334, the campaign uses fake banking, invoice, and business documents, to trick victims into installing malware which, in turn, deploys a malicious extension in Chrome and Edge browsers.

The researchers named the malware “Kremlin”, and say it can steal browser credentials, cookies, session information, monitor browser activity, grab screenshots, and steal information from websites that the victims visit. But the goal of the campaign is primarily to target Brazilian bank users.

A thousand victims

The malware really makes an effort to hide and persist in the target environment. For example, it first checks to see if it’s in a sandbox and if so - it simply won’t run. If instead it determines that it’s running on a real user’s computer, it will deploy an extension with the name “AVSync System Inc.” in an attempt to trick the victim into thinking they have an antivirus addon running in the browser.

It also doesn’t use a fixed C2 server, but rather stores the information on the Ethereum blockchain, since it’s a lot harder to disrupt the communication between the operators and the infected machines that way.

During their investigation, Elastic researchers were able to take control of a domain that the malware used and discovered that it had infected 1,515 systems. Almost all of them (98%) were located in Brazil. They were also able to register the network canary domain and point it to their webhost, which resulted in the loader assuming it was in a sandbox. This also meant “the infections have not moved past the initial access”, Elastic explained.

The full list of indicators of compromise can be found on this link.

Via The Hacker News

• StarCraft general manager and creative director Dan Hay says the series' stories and characters are "remarkably topical" today
• Hay says StarCraft and the real world are both like a "frontier"
• He adds that while StarCraft is science fiction, its "themes and characters" are very similar to "the headlines in our world"

StarCraft general manager and creative director Dan Hay has compared real-world events to the series' stories and characters, calling them "remarkably topical."

In a sit-down interview with TechRadar Gaming following the reveal of StarCraft, a first-person, open-world shooter slated to arrive in 2030, Hay discussed the CGI trailer that debuted at BlizzCon and explained that the team wanted to portray the "tension" that will come with this new story-driven adventure.

The developer also said that building the "stakes" was just as important; he couldn't dive too much into the narrative details, but Hay touched on the series as a whole and its storylines that explore "imperfect" characters, suggesting the new StarCraft will examine similar topics.

"The way in which you are a hero is that you are going to learn how to become one," Hay said. "The way in which you are a hero is that you are going to learn how to become one. So it's not perfect, right? And that's the thing about StarCraft: a lot of the characters are not perfect; they're imperfect, and even the villains are.

"They have an idea of what you think they are at the beginning of the game, and then by the end of the game, you realize, oh, they were something completely different. And I think what's powerful about StarCraft."

Hay continued, saying that when you look at StarCraft's stories and the characters you meet, they're "remarkably topical" to the events unfolding in the real world.

"When you think about our world today, and you think about what is StarCraft? It's a bit of a frontier," he said. "It's obviously a frontier. There's asymmetry in terms of factions that don't agree with each other. There's brinkmanship going on where one faction is being used against another. [In] a lot of cases, it's just over resources.

"A lot of the cases there's politics, and there's intrigue, and there are people being left behind and betrayed, and when you look at that, and then you look at the headlines in our world, it's not super far [apart] in terms of how it feels sometimes. So you feel that tension, and you felt that tension in the teaser [trailer]."

Hay concluded: "I think what's remarkable about StarCraft is that it feels real, and it feels like it translates quite easily into today. And even though it's science fiction, and even though it has this sense of being in this other universe, it feels very close in terms of themes and characters and moments that you could experience today."

• The People v Gorilla Grodd is the only DCU TV show in active development
• HBO's boss recently confirmed as much at the 2026 Emmys
• His comments have got me fearing for Lanterns' future

The boss of HBO has revealed that only one DC Universe (DCU) TV show is officially in the works — and now I'm afraid for the future of Lanterns.

Speaking to Deadline, Casey Bloys confirmed that The People v Gorilla Grodd (TPvGG), aka the recently announced Superman TV spin-off, is the only TV series in active development at DC Studios.

Asked for a status update on DC's myriad television productions, Bloys replied: "People read about development, and they assume that those are greenlit shows and they're [already in] development. And everybody knows that development is just that — you're trying things, you take big swings, [and] not everything comes through.

"I would say our track record in DC, from Watchmen to Peacemaker to Penguin and now Lanterns, is pretty good. We're very selective. We're careful about what we do, nothing has changed. You'll read about development, but focus on what we're greenlighting."

Pressed on whether any other show was in active development, Bloys added: "No, not right now, obviously you've got [The People v] Gorilla Grodd, which I'm very happy about, [they're] doing something different, and I think it's a really fun idea. They just started shooting."

Why I'm fearing the worst for the fate of HBO's Green Lantern TV show

Fans have fallen in love with the abrasive Hal Jordan since Lanterns premiered in mid-August (Image credit: DC Studios/HBO Max)

For fans who've immersed themselves in the DCU since day one, Bloys' comments will be of concern.

Indeed, when DC Studios co-head James Gunn revealed the official lineup for DCU Chapter One in January 2023, six small-screen projects were unveiled as part of the rebooted cinematic franchise's initial slate.

Since then, four more series have been added to the list of potential TV candidates, but only one — The People v Gorilla Grodd — has been greenlit. Meanwhile, some shows that were named as part of Gunn's initial announcement, including Waller and Paradise Lost, have reportedly been shelved. Others, such as Booster Gold, have been trapped in development hell.

John Stewart has also won a whole new army of fans in HBO's Green Lanterns TV show (Image credit: HBO)

With Bloys outright confirming TPvGG is all that DC Studios' TV division is actively working on, it's not surprising that fans — myself included — are fearing for Lanterns' future.

The Aaron Pierre and Kyle Chandler-led sci-fi crime drama, which has seen its audience share seemingly grow week-on-week amid positive word of mouth, is one of the most popular shows on television right now. With three more episodes of its debut season yet to air (at the time of publication, anyway), I'm convinced that its viewership will increase in the weeks to come, too.

Considering that HBO's Green Lantern TV show has been such a hit, it's inevitable that a second season will be made, right? Not if we believe Bloys. After all, there are numerous examples of hugely successful shows — on HBO, HBO Max, and otherwise — that haven't returned for another round. There's no guarantee, then, that Lanterns season 2 will be a thing.

That said, DC Studios has assembled a writers room for Lanterns season 2 to kick around potential story ideas. That doesn't mean DC and HBO have pledged to make another season of the DCU Chapter One program, but it does offer hope that, if the right story angle can be found, another eight-part installment might be *ahem* constructed.

As I said at the start of this article, though, I am worried for Lanterns' fate. Bloys is likely playing it coy, but I'm of the opinion that he would've had something positive to say about Lanterns' future if a sophomore season was more likely to happen than not. Right now, then, I'm reluctantly confident that it'll be a one and done scenario.

Mind you, that doesn't mean we won't see Pierre's John Stewart — and, potentially, Chandler's Hal Jordan — again in the DCU. Pierre's Stewart is already set to play a role in Man of Tomorrow, aka next July's follow-up to Gunn's Superman sequel. Meanwhile, there are unsubstantiated rumors floating around that a DCU Green Lantern movie might be fast-tracked because of how popular Lanterns has proven to be.

If, and it's a big if, we get a big-screen Lanterns follow-up to the hit comic book-inspired show, you certainly won't hear me complaining about the decision not to make a second season.

For now, find out everything we know so far about Man of Tomorrow, get the full lowdown on The People v Gorilla Grodd, and learn all there is to know about Clayface.

On Wednesday, the Senate committee on health holds a confirmation hearing for three top health officials, including surgeon general. Dr. Nicole Saphier is President Trump's third attempt to nominate a surgeon general.

(Image credit: John Lamparski)

A European Union-backed court convicted former Kosovo President Hashim Thaci of war crimes committed during his country's war for independence from Serbia and sentenced him to 25 years in prison.

(Image credit: Koen van Weel)

Here’s my take after testing.

• The Boox Palma 3 has been announced
• This is a phone-sized ereader with a 6.13-inch screen
• Key upgrades include Android 16, stylus support, and a more premium build

If you’re in the market for a phone-sized ereader, you now have another option to consider, as the Boox Palma 3 has just been unveiled — and this successor to the Onyx Boox Palma 2 has some of the key features found on the rival ViWoods AiPaper Reader.

For one thing, just like the ViWoods device, the Boox Palma 3 now has a premium build, with an anodized aluminum frame and a textured fiberglass backplate in place of its plasticky predecessor.

And it’s also had an Android upgrade, launching with a ViWoods AiPaper Reader-matching Android 16, which is up from Android 13 on the Boox Palma 2 and Android 15 on the Onyx Boox Palma 2 Pro.

For me, that’s particularly important, as Android versions don’t tend to get updated on these sorts of devices, so you’re typically stuck with whatever software they launch with; old versions of software can mean security issues, and app incompatibility problems can crop up sooner.

Beyond that, the Boox Palma 3 also benefits from active stylus support, so you can use it with the Boox InkSense Plus stylus. The AiPaper Reader doesn’t have stylus support, nor does it come with a camera, whereas here you get a 16MP one, so those are two potential advantages of the Boox Palma 3.

Personally, I’m not sure I’d want to use a stylus on a device this small, nor do I have much use for a camera on an ereader — but it could be handy for scanning documents, if you’re so inclined.

(Image credit: Onyx)Similar specs and a heavier build

That camera and the rest of the specs, though, are largely in line with those of the Boox Palma 2, including a 6.13-inch 824 x 1648 e-paper touchscreen with 300 pixels per inch, 6GB of RAM, 128GB of storage, a microSD card slot, a front light (supporting both warm and cold color tones), a 3,950mAh battery, a USB-C port, a fingerprint sensor, a speaker, and microphones.

But despite being marginally smaller than the Boox Palma 2 at 159 x 78 x 7.8mm, the Palma 3 is actually 25g heavier at 195g.

That also makes it heavier than the 138g ViWoods AiPaper Reader and the 170g Boox Palma 2 Pro — the latter of which additionally gives you a color screen and an extra 2GB of RAM.

So, this may not be a complete home run — or a total upgrade for anyone who owns the Palma 2 Pro.

Indeed, the Palma 3 is actually priced lower than the Palma 2 Pro, coming in at $339.99 (around £250 / AU$475), though a launch discount softens that price to an even lower $319.99 (roughly £235 / AU$450).

There’s no confirmed pricing for other regions yet, but for reference, while the Palma 2 has been discontinued, the Palma 2 Pro sells for $399.99 / $379.99 / AU$679, and its rival, the ViWoods AiPaper Reader, retails for $279 / £254.99 / AU$399.12.

You also can’t actually buy the Boox Palma 3 yet, but the company lists it as coming soon.

Saudi Arabia accused Yemen's Iran-backed Houthi rebels overnight of trying to attack Islam's holiest city, prompting condemnation from countries in the region but no sign of military support.

(Image credit: Marzouq Al Jabri)

The new iOS 27 feature lets you bounce between two iPhones easily, but the carrier is charging for it. And AT&T is staying hands-free for now.

One of the country's first known female tattoo artists was born on a Kansas farm. Maud Stevens Wagner ran off to join the circus and later became a tattoo legend.

(Image credit: Library Of Congress Prints And Photographs Division)

Federal data shows many borrowers who aren't paying their loans went to private, for-profit colleges — schools that often market to low-income students and rely on federal financial aid for revenue.

(Image credit: Jessica Pons for NPR)

Data suggests many candidates are embracing AI this year to boost their image and portray their opponents unfavorably. But whether it will turn out voters remains unclear.

(Image credit: AI-generated campaign ads)

Because of a data lag, the current reality of access may be even worse.

(Image credit: Anna Watts for NPR)

The Fed is widely expected to raise its benchmark interest rate to combat stubborn inflation. That could make it more expensive to borrow money to buy a car or carry a balance on a credit card.

(Image credit: Win McNamee)

In the span of just over two weeks this summer, three of the world's most closely watched AI developers admitted the same uncomfortable thing. Their own models broke out of the sandbox and touched systems they were never supposed to interact with.

OpenAI disclosed on July 21 that models it was evaluating exploited a vulnerability and compromised production infrastructure at Hugging Face, an incident the company said was driven end-to-end by an autonomous agent with no human directing it.

Days later, Anthropic said three of its Claude models, including Opus 4.7 and its newest Mythos 5, had accessed and compromised the systems of three outside organizations during cybersecurity testing exercises, after a misconfiguration left the models connected to the open internet when they had been told they weren't.

And on August 5, Meta confirmed its Muse Spark 1.1 model breached an unnamed company's systems under strikingly similar circumstances.

A pattern, not an anomaly

At the current pace, this isn't a rare event security teams can plan around once a year. It's becoming a recurring line item. Notably, Anthropic and Meta's incidents traced back to the same third-party evaluation partner, and in Meta's case, the model's cyber risk had already been assessed as no higher than moderate before the very testing process meant to confirm that assessment ended up breaching a real company.

That detail matters as it shows the failure point isn't just the model. It's the surrounding scaffolding of evaluations, permissions, and network paths that organizations assume is contained until it isn't.

This should be viewed as an early warning for organizations about autonomous systems moving from content generation into action execution. The practical lesson, now repeated three times over, is that advanced AI systems can behave in harmful or unexpected ways even when the original goal is not malicious, especially when they are given tools, network paths, credentials, and incentives to complete a task at any cost.

For companies, the takeaway is not to halt AI adoption. It's to treat agentic AI as a new class of privileged workload that requires containment, observability, and enforceable runtime controls.

Govern agents like high-risk digital workers

That starts with AI agent identity management. Companies should double down on this discipline and be very deliberate about what agents are allowed to access and do. Each agent should have a unique identity, scoped permissions, short-lived credentials, and clear ownership, so organizations can trace actions back to a specific system, use case, and accountable business owner.

Access should be limited by default, with explicit approval gates for higher-risk activities such as internet access, code execution, credential retrieval, data movement, or changes to production systems.

In practical terms, organizations should govern AI agents like high-risk digital workers: least privilege by default, separation between test and production environments, detailed logging of tool use and system interactions, and a kill switch that security teams can trigger the moment behavior deviates from policy.

Prevention, monitoring, and the road ahead

Prevention also requires moving beyond traditional application security testing. Organizations should red-team agents against realistic misuse paths, including prompt injection, tool abuse, lateral movement, credential harvesting, data exfiltration, and attempts to bypass sandbox restrictions. They should also continuously monitor agents for harmful impacts, not just technical failures.

That means watching for unauthorized access attempts, unusual tool-chaining behavior, unexpected data movement, policy violations, and actions that could create operational, security, privacy, or reputational harm. Periodic audits should review agent permissions, identities, logs, business justification, and actual behavior to confirm that each agent is still operating within its intended purpose and risk tolerance.

Will this become a trend? With three disclosures in seventeen days, that question is close to settled. Autonomous agents will increasingly be able to discover, combine, and exploit weaknesses faster than traditional security processes can respond.

The risk is not simply “AI hacking AI.” It's autonomous decision-making operating inside complex digital ecosystems where one model, plugin, dataset, API, or identity path can become the bridge into another environment, exactly what played out at Hugging Face, inside Anthropic's testing environment, and now at Meta's.

The companies that will be best positioned are those that pair AI innovation with disciplined identity management, access limitation, continuous monitoring, and routine audit practices, rather than treating each new disclosure as an isolated incident to react to after the fact.

The pragmatic message for executives, especially as this list of companies keeps growing, is that agentic AI can create significant business value, but only if autonomy is matched with accountability, containment, and operational guardrails.

We've featured the best endpoint protection software.

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