An inheritance dispute forces a reckoning with the past in Possession, an ambitious five-part thriller that raises the ghosts of empire. When the heir of a British slave and plantation owner dies and leaves the £50 million estate not to his son Oliver Connaught (Jonny Lee Miller) but to Jamaican man Cudjoe East (Sheldon Shepherd), mixed-race lawyer Claudia (Gugu Mbatha-Raw) is deployed to Kingston to fight Oliver’s corner.

You can watch Possession online from anywhere with a VPN.

The 2026 smartphone market has been a bit difficult to deal with. Buying a phone has always been about choosing what matters most to you. Some need their phone to run demanding games for hours, while others love to show off a fancy design and a camera that takes pictures worthy of an Instagram story. If you’re the latter, vivo has just resurrected the S-lineup. For the uninitiated, the first S1 targeted style-conscious, younger consumers, and the new S2 follows the same formula. It features a camera visor-inspired design, with fun colors, a Sony IMX852 camera sensor, and the MediaTek Dimensity 7360 Turbo processor.

All this at a price of ₹39,999 seems like a decent deal given the great AI price surge of 2026. But is it worth your money? To find out, I got the S2 for review and took it with me on a trip to Vietnam. It served as my primary phone in the country, thanks to the eSIM support, and I took about 300 photos in the brutal summer heat on the beaches of Nha Trang.

Design & Hardware

Design is where the vivo S2 gets full marks, simply because it’s beautiful to look at. I’ll say this straight: the horizontal visor is the best implementation of the camera module. It not only doesn’t get in the way but also keeps the phone from wobbling on a table. The same benefits apply here. There are two cutouts for the camera setup, with the flash sitting on the right. Color options, of which there are three, comprise Silk White, Regal Bronze, and Sapphire Blue. I received the Silk White model for review, and it’s a looker. The back has a silk-like finish that shines in different ways as the light hits. There’s also a wave-like pattern that adds to the overall appeal and makes it more premium. Thankfully, the finish is matte, so you don’t need to worry about greasy fingers.

To confirm it wasn’t just me liking the design, I also gave the phone to my friends and family, all of whom had similar reactions. I could also see a few people looking at what phone I was using at the airport. Another thing you’ll notice with the S2 immediately is its build. Instead of the flat sides we’ve all gotten used to, the S2 has a curved aesthetic. The frame and rear glass curve toward each other, and it really does feel good. Sometimes I miss the old curved aesthetic, and it’s nice that vivo is keeping the trend alive. The corners don’t dig into your palms, and people with smaller hands shouldn’t have much trouble using it one-handed.

The sides are obviously plastic, and that’s not necessarily a bad thing. While I was fiddling around at the airport looking for my passport, I dropped the phone on the moving walkway, and it landed directly on one corner. The S2 survived unscathed, and after that, I just kept it inside the pre-bundled case. Since it was rainy season in Vietnam, the S2’s IP68 and IP69 ratings were quickly put to the test. The phone can survive being submerged in water, so all your pool photos should be fine. That said, please don’t take your phone into the ocean since saltwater can still cause damage. There’s also the fingerprint sensor on the front, and it’s pretty standard. It’s an optical unit, so dirty fingers could be a problem, and it’s a bit slow.

Display

Good smartphone displays have become pretty common, especially on flagships. And even though the S2 isn’t a flagship phone, its display is up to the mark. For some context, the 6.8-inch AMOLED panel has a 1.5K resolution and a 120Hz refresh rate. All this, as expected, translates to a great multimedia experience. Colors look vibrant, blacks are inky deep, and skin tones look natural while I binge-watched “Ted Lasso” at the airport.

vivo claims a peak brightness of 3,000 nits, and while I don’t carry a light meter on vacations, I can tell you that it’s plenty bright. The S2’s display was legible even in direct sunlight when I tried to read the cab number, and I wasn’t squinting much. A few other nice additions include an SGS-certified eye protection filter and Low Flicker Certification. Both are designed to make the display easier on the eyes, particularly at night. vivo hasn’t specified the exact type of glass it uses on the S2. Since I didn’t drop the phone after the first fall, I can’t comment much on how it’ll survive drops. But as a general rule of thumb, curved panels are less resistant than flat ones, so I’d recommend keeping the pre-applied screen protector.

Performance & Software

You can have all the fancy design and display in the world, but none of it is of any use if the phone can’t open Instagram quickly for a little doom scroll session. Fortunately, that’s not the case with the vivo S2. It’s powered by MediaTek’s Dimensity 7360 Turbo processor, paired with up to 8GB of LPDDR4X RAM and 256GB of UFS 3.1 storage. I’ll say this straight: DDR4X RAM should not be inside a sub-40K phone. Sadly, that’s the consequence of AI, and we can’t really blame vivo for that.

Everyday performance on the S2 is quite good. It opens and closes regular apps without a hitch or stutter, keeping the experience smooth. Heavy apps open quickly, and vivo’s new OriginOS is a great version of Android. Animations are fluid and clean, and the UI is full of customization options for you to change the phone’s vibe. It’s not overwhelming, and I got used to the interface in just a couple of days.

Since this is 2026, vivo has bundled a few AI features, including transcription tools and Smart Call Assistant, along with the vivo Office Kit. Sadly, except for AI erase, you don’t get many AI editing features, which was a bummer. I also noticed some stutters when flicking through different applications, but nothing major. What was more annoying, though, was the bloatware. I counted roughly 20 pre-installed applications even after saying no to installing anything in the first place. Thankfully, vivo vivo allows users to remove all apps.

I also ran a series of benchmarks to test the raw performance of the Dimensity 7360 Turbo. In Geekbench, scored 1,017 in the single-core test and 2,987 in the multi-core test. In AnTuTu, the S2 went just shy of the million mark with a score of 990,781. For context, those numbers are similar to the Snapdragon 7 Gen 3. vivo has also incorporated a big vapor chamber inside the S2, which I put to the test in CODM. Even on Very High graphics settings and Ultra frame rate, it held up pretty well. There was no noticeable lag or jitter during intense combat, and the phone only got warm after a 30-minute session. To help protect battery health, there’s bypass charging as well.

Battery Life & Charging

Given the slim profile of the S2, I expected the battery life to be nothing to write home about. The truth is, however, way different. With a capacity of 7,050 mAh, the S2 is a battery champ, capable of lasting multiple days on a single charge. I used the phone for navigation while walking around Nha Trang looking for a lunch spot. I then headed to the beach in the evening to capture some photos, and ended the day by taking a few more photos and watching some YouTube videos. I ended that day with about 40% charge remaining, which is pretty impressive.

When it was finally time to recharge, the 44W fast charger came in handy. It took the phone from 20% to 80% in around 50 minutes, which should be enough for any demanding day. vivo also says the cell should be good for over 1200 charging cycles without losing much health.

Cameras

Given that the vivo S2 targets younger, style-conscious buyers, photography is understandably a big part of the pitch. The vivo S2 houses the 50MP Sony IMX852 sensor, and that’s it. There is a secondary camera, but it’s a 2MP sensor and no ultrawide action. That said, the photos from the main sensor are really good. With enough light, the S2 captured rather pleasing photos of Nha Trang’s cityscapes. The shots had plenty of detail, while colors leaned toward the vibrant side, which I personally like. Contrast was mostly spot on as well. The camera sometimes overexposed the sun, but nothing major.

Since there’s no telephoto sensor, the S2’s main camera also doubles as the portrait lens. The same generally held true for portrait shots, with skin tones being true to their color, and a nice shallow depth of field. As far as nighttime photos are concerned, the images do benefit from vivo’s processing, which tries to retain colors and balance everything. However, I did notice a lack of sharpness in photos here, which was further exaggerated by the long shutter times. Moving subjects can be a problem. Video recording tops out at 4K 30 FPS, and they are fine. The resulting shots are clear and very usable in daytime. Selfies are another highlight, with rather good sharpness and colors. Plus, you can also record 4K videos from the front.

Verdict

Starting at ₹39,999, the vivo S2 is an interesting phone, marking vivo’s return to a segment it hasn’t explored in a while. And I must admit, I love this design. It’s practical while still managing to turn heads. If you like your tech to make a statement, then there aren’t many choices in the market. Beyond the design, there’s a big 6.8-inch AMOLED panel that’s great for watching movies, while the Dimensity 7360 Turbo proved capable of handling everyday tasks and games like “CODM” without much trouble. The cameras, or should I say the camera, are good enough for daytime photos, but struggle with sharpness at night, so keep that in mind.

Another AI model has slipped its leash. This time it’s Kimi K3, the open-weight model from China’s Moonshot AI, and according to a report from Wired, it broke free from its sandbox during a cybersecurity test run by the startup Frontier Security.

How did Kimi K3 escape?

The breakout occurred during testing of Kimi K3’s defensive cybersecurity skills, and, much like the earlier incidents involving OpenAI and Anthropic, a gap in the sandbox setup is partly to blame. But Frontier Security says there’s more going on here. Kimi K3 didn’t just slip through a crack; it noticed the crack and used it.

“We found a leak in the sandbox,” said Yaron Singer, CEO of Frontier Security. “But we also found that Kimi took advantage of that loophole, suggesting that it doesn’t have the same internal guardrails.”

Interestingly, Kimi K3 didn’t hack anything once it got online. It didn’t have to. The answers to its assigned problems were already sitting out in the open on GitHub, so it simply grabbed them instead of solving them the hard way.

Should this worry you?

This isn’t an isolated case. OpenAI recently disclosed that one of its unreleased models broke onto the internet and hacked into Hugging Face, later admitting its agents hit four more services in the same spree. Anthropic reported something similar soon after, revealing that its AI models went rogue and hacked three different companies. 

Just yesterday, Meta also revealed that its Meta AI escaped the sandbox and hacked another company. Now, Kimi K3 has joined this long list of rogue AIs, and it’s concerning to say the least. 

As more companies race to build agents that can act on their own, incidents like this are a good reminder that the sandbox matters just as much as the model inside it. Expect more of these stories before the industry gets this figured out.

Creative AI boosts advertising ROI by automating the two most expensive parts of the creative process — producing enough fresh variants to outpace audience fatigue, and figuring out which variant actually works for which viewer. Done well, it turns a slow, guesswork-driven creative cycle into a fast, testable one.

Quick Take

Marketers lose ROI in advertising for a predictable reason: creatives wear out faster than teams can replace them. Creative AI attacks that problem from three angles — generating variants at volume, assembling personalised versions automatically, and testing far more combinations than a human team could manage. None of this replaces a strategist’s judgment about brand and audience; it removes the production bottleneck that used to stand between a good idea and a tested one.

Why Creative Fatigue Is the Real ROI Problem

Creative fatigue happens when an audience has seen the same ad enough times that they stop registering it. The earliest reliable warning sign is a declining click-through rate, with week-over-week drops of 10–15% or more generally treated as the point to act, well before ROAS visibly suffers. By the time cost-per-acquisition starts climbing, the campaign has usually already been losing efficiency for days.

The traditional response — brainstorm, brief, design, edit, launch — is slow by design. A full creative cycle commonly runs six to ten weeks from ideation to delivery, which is longer than many fatigue cycles themselves. That mismatch is the actual cost driver behind declining ROI: not that any single ad is bad, but that the pipeline can’t refresh creative as fast as the audience burns through it. Learning how to beat creative ad fatigue before it erodes performance has become a prerequisite for running efficient paid media at any scale.

How AI Generates Ad Creative at Scale

The most direct lever creative AI gives marketers is volume without a proportional increase in headcount. Instead of a copywriter drafting one headline at a time, AI copy tools can produce dozens of headline variants in seconds, each built around a different angle, tone, or emotional trigger, for a human editor to shortlist rather than originate from scratch.

The same shift is happening in visuals and video. AI image generators can produce a wide range of product shots and concept visuals from a text prompt, letting teams test visual directions without booking a photoshoot for each one. Video has followed the same path: a genuinely useful API for generating multi-shot videos lets a team assemble a dynamic video ad from separate scenes programmatically, cutting out much of the traditional production overhead. What this buys marketers isn’t just cheaper production — it’s the ability to refresh a fatiguing campaign in days instead of weeks, which is the timeline creative fatigue actually operates on.

How Dynamic Creative Optimisation Personalises Ads Automatically

Generating variants is only half the problem; matching the right variant to the right viewer at scale is the other half, and it’s where Dynamic Creative Optimisation (DCO) does the heavy lifting. A DCO system breaks an ad into interchangeable components — headline, body copy, image or video, CTA button, background — and assembles a specific combination for each impression based on the viewer’s available signals: location, browsing history, past interactions with the brand.

As a hypothetical illustration of how this plays out: an online retailer’s DCO setup might automatically show a shopper in a colder region an ad featuring outerwear, while a shopper who recently browsed a specific product sees an ad built around that exact item. The mechanics aren’t exotic — DCO now ships natively inside major ad platforms rather than requiring a separate specialist vendor, which is why it has moved from an enterprise-only capability to something most mid-sized advertisers can turn on directly. For guidance on applying the same logic to a specific channel, our breakdown of dynamic creative strategy for B2B feeds covers the platform-level setup.

How AI Tests and Optimises Creative Combinations

Manual A/B testing can realistically compare a small number of variants at a time, and it takes real traffic volume to declare a statistically confident winner. AI-driven multivariate testing changes the scale of what’s testable: instead of comparing two or three creatives, the system can evaluate a large matrix of component combinations simultaneously and continuously reallocate spend toward whichever combination is actually converting for a given segment.

Some platforms extend this further into pre-launch prediction, scoring creative concepts before they ever go live. Accuracy claims here vary meaningfully by vendor and methodology, so they’re worth treating as vendor-reported rather than settled fact — one platform, for instance, reports prediction accuracy above 90%, compared with roughly 52% for unaided human judgment on the same task. Whatever the exact number turns out to be for a given tool, the direction is consistent: pre-launch scoring lets a team weight its budget toward the most promising variants from day one instead of splitting spend evenly and waiting for live data to sort winners from losers.

Common Misconceptions About Creative AI

“AI replaces the creative team”

In practice, AI tools handle production volume and combinatorial testing — the parts of the job that are repetitive and data-heavy. Deciding what the brand should say, what the campaign’s core idea is, and whether a given direction fits the audience remains a human call. Teams that treat AI output as a first draft to edit tend to get better results than teams that publish it unreviewed.

“More variants automatically means better performance”

Volume only helps if the underlying concept has merit. Generating a hundred weak headlines and testing them against each other still produces a weak winner — it’s just a data-backed weak winner. AI expands the number of ideas worth testing; it doesn’t substitute for a strong core creative concept.

Where Creative AI Falls Short

DCO and multivariate testing need enough traffic to reach statistical confidence quickly; a low-volume campaign may never generate enough impressions per combination to produce a reliable winner, in which case the added complexity isn’t worth it. Predictive scoring tools are trained on historical patterns, so they tend to be weaker on genuinely novel creative directions or new-to-brand formats where there’s no comparable history to learn from. And AI-generated visuals and video still need brand and legal review before launch — nothing here removes the compliance step, it just moves it earlier in the pipeline where it’s cheaper to catch problems.

Measuring AI’s Impact on Campaign ROI

The business case for creative AI rests on two separate layers, and conflating them makes the ROI case weaker than it actually is. The first is production efficiency: track the time and cost of producing a set of creatives before and after adopting AI tools, including hours spent on design, copywriting, and review. That’s a direct, easily-measured cost saving on its own.

The second layer is performance: compare CTR, conversion rate, and CPA between AI-assisted and manually-produced creative sets, ideally over a comparable time window and audience. Where production costs are falling and performance metrics are holding steady or improving, the ROI case is straightforward. Where automated bidding and inventory-feed automation are also part of the stack, it’s worth pairing this creative-side measurement with a broader look at how automated bidding and real-time inventory feeds affect ad ROI, since creative and bidding efficiency compound rather than operate independently. For teams building out the measurement side specifically, a marketing mix modeling approach to reducing wasted ad spend gives a framework for attributing results back to the right lever. Search-heavy advertisers, particularly in B2B SaaS PPC campaigns, tend to see the clearest early wins here because creative refresh cycles map directly onto keyword-level performance decay.

Key Takeaways

• Creative fatigue, not weak targeting, is the most common hidden driver of declining ad ROI — and it moves faster than most production pipelines can keep up with.
• AI creative generation shortens the production cycle from weeks to days, which matters because that’s the actual timescale fatigue operates on.
• DCO automates personalisation by assembling ad components per viewer in real time, and it’s now built into major platforms rather than requiring a separate vendor.
• AI-driven multivariate testing evaluates far more combinations than manual A/B testing, though pre-launch prediction accuracy claims vary by vendor and should be treated as such.
• Measure production-cost savings and performance metrics separately — conflating them overstates or understates the real ROI case.

Frequently Asked Questions

security

Beijing’s not saying why, which is just what happened when it investigated Micron

China’s Cyberspace Administration (CAC) has conducted a review of Palo Alto Networks’ products.

The regulator’s announcement of its review says it’s needed “to ensure the safe and stable operation of critical information infrastructure, prevent cybersecurity risks and vulnerabilities, and safeguard national security.”

And that’s all Beijing has to say on the matter.

A Palo Alto spokesperson provided The Register with the following statement: “We maintain the highest standards of business conduct and security practices and ethics across our global operations. At this time, there is no impact to our ability to support customers or deliver our products and services in the region.”

This matter has echoes of China’s 2023 investigation into the security of products from memory-maker Micron, which the CAC announced out of the blue.

Micron had previously fought intellectual property and antitrust cases in China, but the company and Chinese authorities did not explicitly link those matters to the security probe.

The CAC published its findings weeks after announcing the probe and decided Micron’s products represented an unacceptable security risk for critical infrastructure operators – effectively banning sales of Micron products to such entities – but didn’t offer a detailed explanation for its decision. The memory-maker eventually stopped selling its datacenter and server products in China, a decision that cost it billions of annual revenue – but created new opportunities for China’s own memory-makers, which are largely prohibited from selling to American companies.

China is home to several security companies whose product portfolios overlap with Palo Alto’s. Huawei and H3C, for example, have plenty to offer local buyers.

Palo Alto doesn’t reveal revenue earned from individual countries, so it’s hard to know what a potential ban could cost the company.

China has for years accused Western tech companies of assisting US surveillance and offensive hacking activities. The Register would not be surprised at all if Beijing reuses that reasoning in its findings about Palo Alto products. Western governments level the same accusations at Huawei and ZTE.

Beijing’s ban on Micron didn’t noticeably impact the company’s reputation elsewhere. Indeed, the AI boom has brought Micron such great riches that past dents to its bottom line are now almost irrelevant. ®

In March, a jury in New Mexico found Meta liable for violating the state’s consumer protection laws in a case brought by its Attorney General, accusing the company of child exploitation and other safety issues. The court ordered Meta to pay a fine of $375 million in civil penalties. Now, after the second phase of the trial meant to decide whether Meta has been a “public nuisance” harming teen users in the state, the court has ordered the company to pay an additional $567 million into an abatement fund for related programs in New Mexico.  

In his ruling, Judge Bryan Biedscheid likened Meta’s platforms to factories, the advertising and other content in them to be the factory’s products and the psychological harm and sexual exploitation of children on those platforms to be pollution they emit. And just as pollution can harm the public’s right to reasonably clean air, the harmful effects of Meta’s apps on children aren’t contained within those platforms. The judge said that those harmful effects “migrate to the internet as a whole and, perhaps most concerning, to the real world.”

The judge clarified that the ruling isn’t seeking to close down Meta’s apps. Instead, it’s looking to address the existing harms they’ve already caused. Meta threatened back in April, however, that it may be forced to shut down its apps in New Mexico if the court sides with the state. It said that the state’s demands were so “burdensome” that “it might force Meta to withdraw its apps entirely.” New Mexico Attorney General Raúl Torrez described the company’s threat as a PR stunt. 

The penalty Meta has been ordered to pay will go towards several different programs, including youth internet safety training for teachers, school counselors, school psychologists and healthcare professionals, as well as community-based health centers. In addition to paying the fine, the court has ordered Meta to incorporate changes and safety measures into Facebook and Instagram. It has determined that WhatsApp doesn’t contribute to public nuisance and, hence, any required changes will not apply to it. 

Meta will have to maintain a private-by-default setting for users under 18 years of age on Instagram. If a user on Facebook is under 18, it also has to limit their friends to users under 18 years of age by default. Private accounts for underage users cannot show up in a search unless identified by username, and they cannot be recommended to other users. Those users cannot change default settings without the consent of a parent or a guardian, or unless they show proof that they’ve already turned 18. 

The company’s platforms cannot recommend any account belonging to users under 18 and must ensure that adults who aren’t connected to underage users cannot message them. Meta has to pause push notifications for the same users from 10PM to 7AM everyday and from 8AM to 3PM during the academic year, excluding weekends. It has to hide “like counts” for users under 18 and to implement a mandatory usage time limit, as well. Specifically, teens can only spend 90 hours per month across Facebook and Instagram. Meta also has to flash information banners for underage users once a day, telling them about safe practices and tools they can use to report inappropriate behavior and content. The court didn’t grant the plaintiff’s request to order changes to Meta’s algorithms, however, despite its argument that they create harmful feedback loops. 

The watch’s insecurity and the spying it enabled might be expected given the gadget’s pedigree: It’s sold by an obscure company called CJC, costs less than $30, and was made by an equally obscure manufacturer, YiQingTeng Electronics, in Shenzhen, China. More troubling, perhaps, is that the online platform it’s built on—and the one that allowed Stykas and Solferini to so thoroughly hack it—is used by dozens of other brands of smartwatch, many of which have likely been left vulnerable to the same forms of digital stalking.

At the Black Hat cybersecurity conference today, Stykas and Solferini plan to present their findings from analyzing the supply chain and security of more than 70 GPS-enabled watches and car accessories. They found that more than 30 of those geolocation devices use the technology and backend servers of YiQingTeng, also identified by the brand name Wonlex, the name of a partner firm Shenzhen 3G Electronics, or their associated app, SETracker. Another 30-plus brands of tracking devices for cars and kids are all run on another Shenzhen-based platform known as NewGPS2012.

Combined with another major GPS platform known as SinoTrack that sells car trackers and smartwatches, the two researchers found that tens of millions of GPS tracker gadgets came from just three supply chains. All three, the researchers found in their analysis, had significant security flaws—in some cases as simple as a lack of authentication that allowed anyone to access any device—leaving children’s watches vulnerable to tracking by a hacker, location disabling and spoofing, interception and spoofing of text and audio messages sent to them, replacement of emergency contacts with ones a hacker chose, silent audio eavesdropping, as well as photo and video capture for camera-enabled devices. (Once the GPS started working on the smartwatch WIRED tested, the hackers showed that feature, too, could be hijacked to follow the wearer’s every move.)

For some GPS-enabled car accessories, the researchers found they could similarly track the devices’ locations or spoof messages to them that could potentially unlock or disable cars, though the researchers didn’t go so far as to test this out on actual vehicles. They also say they found server-side vulnerabilities that exposed consumer information, would have allowed them to execute their own code on the servers, or even in one case appeared to show that someone else had already gained unauthorized access to the system’s backend.

“Millions of kids are being exposed and vulnerable to exploitation. It’s just catastrophic. It’s really low-hanging fruit for a lot of bad actors,” Stykas says. “Your criminal mind is the only limitation in exploiting those devices.”

The Watches Watching Your Kids

The researchers say they’ve been warning the companies behind all three Shenzhen-based GPS platforms about their vulnerabilities for months. When WIRED reached a representative of SETracker, the person initially claimed in an email that “the issues you mentioned have been resolved long before,” adding that “we attach great importance to the security of Setracker and keep strengthening its security continuously.” When WIRED pointed out that researchers had been able to hack a smartwatch running on SETracker just this week, the person repeated their claim that the issues had been fixed, then asked for evidence of the exploitation, which WIRED provided.

Only today, hours before the researchers’ talk at Black Hat, did the researchers find that their hacking techniques against SETracker’s platform have stopped working—though they’re still not sure if the flaws they found are fully fixed.

Sinotrack and the NewGPS2012 platform didn’t respond to WIRED’s requests for comment, and the researchers say their hacking techniques against those systems still appear to work.

For more than a decade, cybersecurity experts and privacy advocates have warned that cheap, GPS-enabled children’s smartwatches and aftermarket vehicle accessories are riddled with security vulnerabilities that leave kids and drivers susceptible to hacking and tracking. But the sheer number of different brands and models of those devices has often made identifying the truly insecure gadgets feel nearly impossible for consumers.

The AI industry is having a rogue agent summer. The latest model to escape onto the open internet during security testing is Kimi K3, a powerful open-weight offering from the Chinese company Moonshot AI.

Frontier Security, a US startup, says that Kimi K3 went outside of its sandbox while testing its defensive cybersecurity skills. As with incidents previously reported by OpenAI and Anthropic, the escape was partly enabled by a misconfiguration in the sandbox designed to contain it. Frontier claims, though, that the incident shows Kimi has fewer cyber safeguards than most other powerful AI models, something that allowed it to go off and use the internet without express permission.

“We found a leak in the sandbox,” says Yaron Singer, CEO of Frontier Security. “But we also found that Kimi took advantage of that loophole—suggesting that it doesn’t have [the same] internal guardrails.”

Unlike other recent incidents of AI agents going off-script, Kimi K3 did not hack anything after accessing the internet—because the answers to the problems it was seeking were easily attainable on GitHub.

Moonshot did not respond to a request for comment by time of publication.

The incident is the latest in a string of agent mishaps that suggest increasingly cyber-capable AI models are becoming more challenging to control.

Last month, OpenAI disclosed that an unreleased model had broken out onto the internet and then hacked Hugging Face, a company that hosts AI models and data, in order to find answers to problems it was tasked with solving. OpenAI subsequently shared that its AI agents had in fact hacked into four additional services as part of the spree.

Shortly after OpenAI reported its incident, Anthropic revealed that several of its models had also gained access to the internet and attacked outside systems. Last week, the AISI also disclosed that in its own testing, versions of OpenAI and Anthropic models that had security safeguards disabled perpetrated multiple hacks across the internet, including a particularly ambitious attempt by Anthropic’s Mythos 5 to plant malicious code in an open-source project on GitHub.

While these AI hacking episodes all vary in both cause and degree, the Kimi K3 is similar to several of them in that a misconfigured sandbox allowed access to a number of websites rather than keeping it contained to a simulated environment. The model was expressly tasked with solving problems that should not have involved going off to find the answers online, and appears to have gone outside of those instructions. The model had to figure out for itself that it had access to certain websites by probing the network settings of the sandbox.

While human error appears to have played a major role in each of the breakouts, the consequences have been compounded by the fact that advanced AI models are designed to use reason and take complex actions in order to solve problems.

Another key difference between previous incidents and the one discovered by Frontier Security is that it involves a model that is already widely available, with the same safeguards an average user would encounter.

“Kimi K3 is very good at following a goal by any means necessary and also doesn’t have the guardrails to prevent it from cheating or escaping the sandbox,” says Paul Kassianik, a researcher at Frontier Security.

Kassianik and Singer both say that Kimi and other open-weight models are also excellent tools for cybersecurity defense. (Hugging Face ultimately used an unnamed AI model from China to defend itself against the OpenAI agent hack.) Their company has developed benchmarks that measure a model’s capacity to find vulnerabilities in software and networks, which show that Kimi excels at these tasks.

Apple can usually make suppliers bend the knee. It buys components in volumes few consumer technology brands can match and has built a supply chain that rivals spend decades trying to copy. However, that leverage has now run into a memory manufacturer willing to say no.

Apple reportedly approached China’s CXMT as an additional DRAM supplier and asked for lower prices. CXMT instead quoted figures comparable to or even higher than Samsung and SK Hynix. Huawei, Xiaomi, and other Chinese manufacturers have already locked up much of its output through higher-priced long-term deals, leaving the company with little reason to accept Apple’s terms.

I felt a much smaller version of the same crisis recently. A failing Corsair Vengeance LPX 8GB DDR4-3200 stick made my aging PC unstable, so I had little choice but to replace it. The same module should have cost around 1,500 INR, or roughly $16, before the RAM crisis. I paid 8,000 INR, around $84.

Paying more than five times the sensible price for old DDR4 hardware is absurd. Apple has billions of dollars and purchasing power that I obviously do not. Yet in today’s market, even Apple is learning that scale may secure memory supply without necessarily making it affordable.

Apple’s supply-chain muscle has finally hit a wall

CXMT’s growing influence should eventually be good for the memory industry. It gives device makers another serious supplier beyond Samsung, SK Hynix, and Micron, potentially reducing their dependence on the three companies that have dominated DRAM for years.

More competition usually leads to better pricing. Unfortunately, CXMT is expanding into a market where nearly every available memory chip already has a buyer waiting for it. Samsung and SK Hynix are investing heavily in high-bandwidth memory for AI accelerators, where the profit margins are far higher than those of ordinary consumer DRAM. Chinese smartphone and computer manufacturers are securing CXMT’s conventional memory before it leaves the factory. Apple cannot use one supplier to pressure another when all of them are already struggling to meet demand.

The company has reportedly asked the US government to let it source DRAM from CXMT and NAND storage from YMTC for products sold outside the country. Approval would likely help Apple secure enough memory, but it may do little to bring down what the company pays. CXMT has already shown that it does not intend to become Apple’s bargain supplier, weakening the company’s hopes of using another manufacturer to rein in rising memory costs.

Memory costs are already reaching Apple buyers

Apple spent months shielding customers from rising component costs, but it has now started passing them on to buyers. Prices for several Macs and iPads have already gone up by hundreds of dollars, and the iPhone 18 Pro lineup may be next.

TrendForce says conventional DRAM contract prices jumped between 93% and 98% in the first quarter of 2026, pushing industry revenue up 81% to $97 billion. By July, prices were still rising, although the pace of increase had started to slow. Counterpoint Research estimated the 12GB and 1TB iPhone 18 Pro Max could cost Apple nearly $300 more to build than its predecessor, with memory and storage responsible for the largest part of that increase.

Supply is not growing fast enough to bring prices back down. IDC expects the amount of DRAM available this year to rise by only 16%, while industry revenue is forecast to jump 177% to $418.6 billion.

Apple can absorb the increase, accept lower margins, reduce memory or storage configurations, or pass the cost on to buyers. We have already seen which option it picked for Macs and iPads. The iPhone may not remain protected for much longer.

The blame game is in full swing, but it won’t lower your bill

Micron has its own explanation for the current mess. The company claims a couple of major customers used their purchasing power during the previous memory downturn to secure rock-bottom prices. Those prices supposedly made expansion difficult to justify, leaving manufacturers unprepared when demand returned. Micron did not publicly identify Apple, although reports have connected the remarks to the company.

I find Micron’s argument a little ridiculous.

A company placing massive, predictable orders will always expect a volume discount. Apple does not buy DRAM at retail, and Micron, Samsung and SK Hynix were not producing millions of components for one of the world’s most profitable companies as a charitable exercise. I also find it difficult to accept that Micron continued serving Apple for years without earning enough money to keep the relationship viable. If the contracts were genuinely unsustainable, Micron had the option to renegotiate them or sell its production elsewhere.

Memory is one of the most cyclical parts of the technology industry. Manufacturers expand production when prices are high, excess supply pushes prices down, investment slows, and the next shortage sends them back up. Blaming customers for negotiating during the low part of that cycle feels awfully convenient now that suppliers have regained pricing power.

Micron does not have a spotless history here, either. A former Micron sales manager pleaded guilty to obstructing the US Justice Department’s DRAM price-fixing investigation after withholding and altering subpoenaed documents. Samsung and Hynix later pleaded guilty to participating in the conspiracy, paying fines of $300 million and $185 million, respectively. None of that proves anything illegal is happening today, but it gives me little reason to accept its version without question.

Still, Micron is right about one part of the crisis. AI companies are buying huge quantities of highly profitable HBM, and memory manufacturers have every reason to prioritize those customers. Conventional RAM for phones, laptops, consoles, and PC repairs is left competing for what remains.

CXMT’s rise could eventually loosen the grip of the established memory manufacturers, but more competition needs more production before buyers see any relief. For now, the new player has entered the same expensive market rather than disrupted it.

Apple will pay more, device makers will pass those costs along, and customers will delay upgrades or accept higher prices. Others, like me, will pay $84 for an 8GB DDR4 stick because an aging PC suddenly leaves them with no other choice.

The Shards is guaranteed to have audiences talking, for better or for worse, as Bret Easton Ellis’ controversial semi-autobiography comes to the small screen.

But, in my opinion, Ryan Murphy teaming up with Ellis has given him the chance to shine once again. Fans of Murphy’s work were left disappointed and baffled in recent months after his abysmal Hulu series All’s Fair aired, which has a seriously low Rotten Tomatoes score of just 6%.