A previously undocumented malware botnet named AryStinger has compromised more than 4,000 outdated routers to turn them into proxies for malicious traffic.

Researchers at Qianxin’s XLab threat intelligence team say that the malware converts infected devices into remotely controlled “executors” that can perform scanning, proxying, tunneling, command execution, and other activities on behalf of the attacker.

“The attacker can split a massive scanning task into multiple small chunks and distribute them to different Executors for parallel execution,” XLab researchers note.

“With this distributed-like design, the attacker can efficiently complete the early “footprinting” activities, thereby providing strong assurance for the smoothness and success rate of subsequent intrusion operations.”

Apart from using compromised routers as a springboard for malicious operations, XLab warns that the malware can also tamper with DNS settings, hijacking the user’s browsing, and silently monitor and potentially steal all inbound and outbound network traffic.

AryStinger exploits older flaws such as CVE-2013-3307, CVE-2016-5681, and CVE-2025-11837, targeting primarily D-Link DIR-850L, D-Link DIR-818LW routers.

The two router models were previously targeted by the AVrecon malware botnet that Lumen communications services provider Lumen disrupted in 2023.

Qianxin’s telemetry data shows that almost half of all infections are located in South Korea (48.5%), followed by China (31.8%), Sweden (6.4%), Malaysia (3.5%), and Singapore (2.5%).

XLab researchers found two variants of the AryStinger malware: a C-based version targeting mostly outdated routers, and a Go-based one that focuses on NAS systems, but currently with a far more limited reach.

The NAS version is the most advanced of the two, featuring additional capabilities such as IP and DNS scanning, command execution, payload execution, and internal network reconnaissance through the integration of open-source penetration testing tools.

The researchers noted that AryStinger’s distributed DNS-scanning infrastructure could potentially be repurposed to generate large volumes of DNS queries against resolvers, although they did not observe any such attacks.

Regarding the NAS version’s code execution capabilities, XLab says there’s support for Shell commands, as well as Go, Java, and Python source code.

However, there are some limitations to using source code instead of compiled binaries, as compilation requires language runtimes on the host, and the process as a whole introduces noise that can break stealth.

The researchers did not attribute AryStinger to any known activity cluster, stating that “many mysteries surrounding AryStinger remain to be solved.”

Owners of end-of-life (EoL) routers should replace them with new, actively supported models, apply the latest available firmware updates, change the default administrator account password, and disable remote management panels.

Security teams log 54% of successful attacks and alert on just 14%. The rest move through your environment unseen.

The Picus whitepaper shows how breach and attack simulation tests your SIEM and EDR rules so threats stop slipping by detection.

Get the whitepaper

Let’s be honest: few iOS design changes have sparked as much debate as Liquid Glass. When Apple first introduced it with iOS 26, the internet immediately split into two camps. Some people loved the fresh, translucent look, while others couldn’t stand it and felt it made parts of the interface harder to read. I happened to be firmly in the first camp. At the time, I was using an iPhone 14 Pro Max, and installing the update was one of the first things I did. I loved how the new design made iOS feel more modern and dynamic. The transparency effects gave the interface a sense of depth, making the entire experience feel fresh again.

That said, it’s easy to understand why not everyone felt the same way. After months of feedback, screenshots, hot takes, and endless debates online, Apple eventually responded by giving users more control. Instead of forcing everyone into the same look, it introduced options that let people choose between a clearer glass effect and a more tinted appearance. With iOS 27, Apple is putting the Liquid Glass debate completely in your hands. A new slider lets you customize the effect exactly the way you want it, whether you prefer a crystal-clear look or something easier on the eyes. Here’s what it does and how to make the most of it on your iPhone.

The slider that settles the debate

The Liquid Glass slider is essentially Apple’s way of letting you decide how much of the new design language you actually want to see. In everyday use, that means you can customize the experience to match your preferences. If you’re like me and enjoy the full Liquid Glass aesthetic, you can keep things more transparent and let your wallpaper shine through menus, widgets, and system panels. It gives iOS a lighter, more layered feel that really shows off the design Apple has been building toward.

On the other hand, if you find all that transparency distracting or simply prefer a cleaner look, you can dial up the tint. Doing so makes interface elements more robust and improves contrast, making text and buttons easier to read at a glance. What I like most is that Apple is no longer treating customization as an all-or-nothing choice. You don’t have to love Liquid Glass exactly as Apple designed it, nor do you have to turn it off entirely. The new slider lets you find a middle ground that works for your eyes and your style, making iOS feel a little more personal.

The road to Liquid Glass starts here

Before you start tweaking the Liquid Glass slider, you’ll need access to iOS 27. I’ve been testing the feature on my iPhone 16e through Apple’s developer beta, which is currently the only way to try it out. If you’re comfortable running beta software, getting started is fairly straightforward. Anyone can enroll in Apple’s developer program for free and gain access to the latest developer builds. Once you’ve signed up, simply open the Settings app, go to General, tap Software Update, and select Beta Updates. Choose the iOS 27 Developer Beta from the list, and the update should appear on your iPhone within a few moments.

That said, beta software isn’t for everyone. Early builds can occasionally be buggy, and some apps may not work exactly as expected. If your iPhone is your primary device and you’d rather avoid the risks of pre-release software, there’s nothing wrong with sitting this one out. The good news is that the wait shouldn’t be too long. Apple will eventually bring the feature to everyone through the public release of iOS 27, so if you’d rather have a more stable experience, it may be worth holding off for a little while longer.

Time to play with the slider

Actually changing the Liquid Glass effect is surprisingly simple. Once your iPhone is running iOS 27, open the Settings app and look for the Appearance section. From there, tap Liquid Glass to access the new customization controls. This is where you’ll find Apple’s new Liquid Glass slider. Unlike the older setup, which limited you to a couple of preset looks, the slider gives you much finer control over the effect. The best part is that you don’t have to guess what each setting does because the interface changes instantly as you adjust it.

I spent a few minutes moving the slider back and forth just to see how different parts of iOS reacted. Sliding it toward one end makes menus, panels, and interface elements appear more transparent, allowing more of your wallpaper and background content to shine through. Moving it in the opposite direction adds more tint and contrast, giving the interface a cleaner and more solid appearance. There’s no right or wrong setting here, which is what makes the feature so useful. You can leave it at either extreme or settle somewhere in the middle if you want a balance between style and readability. As you move the slider, iOS previews the changes in real time, making it easy to find a look that feels right without constantly jumping between menus. My advice to you would be not to rush it. Spend a minute experimenting with different levels of transparency and tint. You might be surprised by how much a small adjustment can change the overall feel of your iPhone.

The best iPhone look is the one you choose

One thing I’ve always appreciated about iPhone is having the ability to make the experience feel a little more personal, and the new Liquid Glass slider fits perfectly into that idea. I’ve been running iOS 27 on my iPhone 16e for over a week now, and while this might seem like a relatively small addition on paper, it’s one of those features that makes a difference every time you pick up your phone. More importantly, it feels like a sign of where Apple is heading with iOS: giving users more flexibility without making the software feel complicated. And from everything I’ve seen so far, that’s a direction I’m happy to see the company take.

There’s still plenty for me to explore in iOS 27, and I’m sure more surprises will reveal themselves over the coming weeks. But if you’re eager to try the new design for yourself, my advice is simple: be patient. The stable release will almost certainly offer a smoother experience than the early beta builds, letting you enjoy all these new features without the usual beta headaches. When that update finally arrives, the Liquid Glass slider is one of the first settings I’d recommend checking out. You might be surprised by how much a small adjustment can change the way your iPhone feels.

AI inference company Baseten is close to finalizing a stunning $1.5 billion funding round at a $13 billion valuation, the Wall Street Journal reports. Just five months ago, the startup announced that it had raised a $300 million Series E at a $5 billion valuation. And that round was just nine months after raising a $150 million Series D.

If finalized, this latest round would represent a 160% increase in valuation in less than half a year. However, the WSJ reports that this is a split-priced round, a tactic startups are using to boost their headline valuation and make lead investors look good on paper. Some investors in this latest funding round are reportedly coming in at a $13 billion valuation, while others at $11 billion, sources told the Journal. This deal is said to be co-led by Spark Capital, Sands Capital, Altimeter Capital, and Wellington Management. 

Launched in 2019, Baseten is a startup benefiting from what The Next Wave hailed the “inference gold rush,” in which VCs are pouring enormous amounts of money into companies building the inference layer. Inference is what the model does after a user submits a prompt. Baseten promises to handle inference quickly while controlling costs by routing requests to the best-for-task model, especially to competent, less-expensive open source alternatives.

Tournament favorites Spain will look to put things right against Saudi Arabia, after both sides opened their respective FIFA World Cup 2026 campaigns with a point. Lamine Yamal hopes to start for La Roja.

Spain play in Atlanta for the second game in a row but the 75,000-capacity stadium will already hold bad memories after La Roja’s goalless draw with World Cup debutants Cape Verde on Monday. Head coach Luis de la Fuente even chucked on star wingers Yamal and Nico Williams in a desperate bid to find a goal, as both players continue their convalescence from recent injuries. The main talking point is now whether the duo will start what feels a must-win game for European champions.

As I watched Graham Sykes climb onto his rocket bike, I was worried for a moment that I was about to film a man as he died. But as he hurtled past me at hundreds of miles per hour and engulfed me in a cloud of steam, I realised I needn’t have worried — this is just a normal day for Sykes. 

I was at the Santa Pod raceway in Bedfordshire, England, meeting Sykes and his team as they prepped his entirely steam-powered bike — dubbed Force of Nature — for a potentially record-breaking speed attempt during a drag-racing festival. After battling through the crowds flooding into the venue, I eventually found Sykes and his team among cars and bikes of all shapes and sizes, diligently preparing Force of Nature for its one scheduled run that day. 

Advertisement

Watch this: World’s Fastest Steam-Powered Rocketbike

05:32

Sykes, who seemed far calmer than I expected, offered me a marshmallow as I got my first glimpse of the bike, “I tend to not eat a great deal before a run, except for sugary sweets — we’ve all got our vices!” he said. 

The bike looked like nothing I’d ever seen before. Long and sleek with enormous funnel-shaped exhausts on the back, the only thing that marked it out as a motorcycle was the fact it was a vehicle on two wheels. A mechanical engineer by trade, Sykes has made almost every component himself from his workshop in his back garden. 

Not that you’d guess. Peering close up at various components, I felt I was looking at something crafted in a NASA lab rather than in someone’s garden shed. At the heart of the bike’s steam propulsion system is a 120-liter boiler, heated by a burner to around 260 degrees Celsius (around 500 Fahrenheit). That boiling process creates an immense amount of pressure inside the tank, which is released when the starting lights go green in about 3 seconds, propelling the bike to speeds of over 200 miles per hour.

The boiler is the only component not built by Sykes He instead sourced it from a company that manufactures pressurized vessels for the nuclear and oil and gas industries. The reason simply comes down to safety. “If it exploded, it wouldn’t just be myself that would be injured or killed,” said Sykes. “It would be everyone else around me too.” 

Despite the very real risks involved, Sykes struck me as very calm and relaxed on the day. He was clearly enjoying himself as he helped the team do the pre-run checks and chatted with excited fans who came to the team’s base to meet Sykes and get his autograph. He was clearly in his element. 

“I always wanted to ride a rocket bike,” he said “But no one was going to ask me ‘Hey Graham, do you fancy riding my rocket bike?’ so the only way to do it was to build one. In the 1970s Evel Knievel tried to jump over Snake River Canyon and that was a super-heated water rocket, so that’s what inspired me.” 

But nerves do still set in, even for Sykes. “Every time you get on the bike, you have trepidation,” he said “You have that adrenaline and you’ve got that little bit of reservation in your head that says ‘when I press this button, my life is gonna change. Hopefully for the better.” 

I positioned myself next to the track, with a clear view of the starting line. I could see Sykes and the team preparing, and had a nice bit of space that would allow me to see him zoom past. I was there to film the spectacle too, but when each run lasts a matter of seconds, it’s not an easy task to capture. Aside from the camera in my hand and the three cameras I had on tripods, I’d also attached a number of GoPro cameras to his bike (using industrial clamps to ensure they didn’t fly off with the force of the acceleration). 

Even so, I was nervous about missing the one opportunity I had to capture Sykes in action. And I was right to be: Before his run, I practiced on other drag races, from tuned-up road cars to hot-rods powered by literal rockets pulled out of fighter jets. The speeds these vehicles achieved were astonishing to me, and the noise was like nothing I’d ever heard. 

Advertisement

But the practice helped me prepare to get the shot, and as I got the thumbs up that the run was soon to begin, I braced myself. As, I imagine, did Sykes. “When we get the bike to the starting line, Diane, my wife, takes the safety pin out, shows me the pin to say that it’s out and the bike is live, taps me on the head which is as good as a kiss, and… off we go,” he told me.

“Nothing can prepare you for what you’re going to experience. It’s like being kicked from behind — the G-Force pulls your body backwards.” 

The lights counted down, turned green and Sykes went off like a bullet. A huge plume of steam erupted from the bike’s exhaust, knocking back one of my cameras, positioned nearby, and sending it hurtling 30 feet into a barrier. I panned my camera quickly as he shot past me, before the wall of steam swept over me. 

It was astonishing to watch and I can’t begin to imagine what it must feel like for Sykes on the bike. I’ve driven some speedy cars in my time at CNET but the fastest acceleration I’ve experienced is 0 to 60 miles per hour in about 2.8 seconds. That felt insanely fast to me — fast enough that I didn’t like it. I felt the edges of my vision blurring and I didn’t want to do it again. 

Sykes does 0 to 60 mph in under half a second. 

The Force of Nature bike didn’t break its own records that day. But the run, at least, was a safe one. “Once I see that parachute come out at the end, I know everything’s all right,” said Diane.

Every run is also a great performance that shows the amassed crowds the true power of what steam can do. “People think that steam is an old-fashioned, out-of-date power source, said Sykes. “But every power station that generates power from fuel is really powered by steam.” 

“When we first started building [the bike] we wanted to run a 5-second quarter mile, with a 200 mph pass — neither had been done before using steam,” he added “We’ve since achieved both of those goals.” 

Sykes and the team hope to achieve a 4-second run in the future. From what I saw of their precision, dedication and passion, I don’t think it’ll be long before they get there. 

Blender is one of the poster children for Open Source Software– proof that something hacked together by enthusiasts could grow to rival the big boys in 3D modeling, animation and rendering after it was abandoned by its original corporate owners. Once you climb that initial learning curve, which can indeed feel cliff-like, you can do almost anything in Blender you can in paid competitors– almost.

Traditionally, one of the weak points has been simulations, with even those working in Blender professionally offloading simulation to programs like Houdini. According to [3Dan], once version 5.2 is out of beta in July, that may become a thing of the past. 

Simulations aren’t a necessary part of a 3D animation software, but they are very, very nice to have. If you want realistic-looking fluids, hair, or cloth, it’s incredibly difficult to animate it by hand. One, because there are so many degrees of freedom in, say, flapping cloth, keyframing is a major pain, but also figuring out how to make the model move and deform realistically is by no means trivial. It’s easier to offload all that on a physics simulation; then, as long as the physics is realistic, the animations will be as well.

That’s not easy, computationally speaking, and one thing that’s clear is there’s been work behind the scenes to optimize the simulation algorithms, not just improve the workflow, as the basic “drop cloth on a monkey head” demo now runs twice as fast. The new workflow itself bring simulations more into line with how Blender has been going– it’s part of geometry nodes now. So there’s simulation nodes you bring in, but that means things like tearing cloth become quite straightforward compared to the occasionally byzantine workarounds required before. This node-based workflow also brings Blender more into line with how paid software works these days.

[Dan] demonstrates the power of it by adding air pressure to a cloth simulation with some custom nodes, inflating and popping a fabric sphere. He also demonstrates how cloth simulation can be applied to animate realistic foliage. This update probably doesn’t have Houdini developer SideFX shaking in their boots, but it might allow some animators to stop paying that license and go fully-open source, which is great to hear.

While the work on the simulations engine is raising the bar on what was, traditionally, one of the weak points of the software, v5.2 brings oodles of improvements across the whole gamut of what blender can do– which is a lot. See them all on the official release notes. Even if you’re not into digital sculpting or animating, you may find yourself downloading a copy of Blender at some point to add texture to 3D prints, or make fancy resin-print miniature models FEM-friendly. The right addon can even let Blender do parametric CAD, if you want open-sorce and can’t stand FreeCAD. Though FreeCAD is getting better all the time, too.

Jio Platforms, the digital and telecom arm of Mukesh Ambani’s Reliance Industries, filed its draft red herring prospectus with India’s securities regulator on Friday for what would be the country’s largest initial public offering. The filing covers a fresh issue of up to 270 million shares, with no offer-for-sale component, meaning every rupee raised flows directly into the company’s balance sheet.

The IPO is expected to raise approximately $3.8 billion, according to people familiar with the matter. That would surpass Hyundai Motor India’s $3.3 billion listing in October 2024, currently the record for an Indian maiden offering.

The DRHP specifies that 275 billion rupees ($2.9 billion) of the net proceeds will go toward prepaying external commercial borrowings held by Reliance Jio Infocomm, its telecom subsidiary. The remaining funds are earmarked for general corporate purposes.

The borrowings in question consist of three loan facilities denominated in dollars and yen, totaling 300.6 billion rupees. Lenders include Australia & New Zealand Banking Group, Bank of America, Barclays, BNP Paribas, and Citibank. All three facilities are scheduled for repayment between March and June 2028, but Jio Platforms intends to prepay them in full or in part from the IPO proceeds.

Ambani announced the filing at Reliance Industries’ 49th annual general meeting on June 19, describing the listing as a step toward unlocking value for shareholders. The IPO will be led by Akash, Isha, and Anant Ambani, the next generation of the family.

Nineteen banks have been appointed as book-running lead managers, including Morgan Stanley, Goldman Sachs, J.P. Morgan, and Kotak Mahindra Capital.

The deleveraging strategy is significant. Jio Platforms’ net debt stood at 275.8 billion rupees as of March 2026, down from 452.7 billion rupees a year earlier and 484.4 billion rupees in March 2024. A successful IPO would eliminate the bulk of the remaining foreign currency exposure and reduce the company’s annual servicing costs.

The company said in the prospectus that repaying the debt would improve its ability to raise future resources for business development and position it for continued investment in 5G network densification, fixed broadband expansion, and AI and cloud services.

Jio Platforms operates through its telecom subsidiary Reliance Jio Infocomm, which is the world’s second-largest mobile operator by single-country subscribers after China Mobile. As of March 2026, it had 524.4 million subscribers, with 268.5 million already on its 5G network, making it the largest single-country 5G operator outside China in a market racing to scale its digital infrastructure.

In the financial year ending March 2026, Jio reported operating revenue of approximately 1.47 trillion rupees ($15.6 billion) and a net profit of roughly 300 billion rupees ($3.2 billion). EBITDA rose 18.8% to 762.6 billion rupees, with margins improving to 51.9%.

At a valuation above $130 billion, which analyst estimates place between $131 billion and $180 billion, the IPO would make Jio Platforms one of the most valuable companies to list in Asia. The offering represents roughly 2.9% of post-issue equity, enabled by a March 2026 regulatory change that allows companies valued above 5 trillion rupees to list with just 2.5% public float.

Meta holds a 9.99% stake and Google holds 7.73%, both acquired during a 2020 fundraising round that brought in more than a dozen global investors including KKR, Vista Equity Partners, Silver Lake, and sovereign wealth funds from Abu Dhabi and Saudi Arabia. The fresh-issue structure means none of these investors are selling in the IPO itself, though the DRHP does not restrict future secondary sales once lock-up periods expire.

The timing places Jio’s filing alongside a broader wave of major Asian tech listings. At the same AGM, Ambani outlined a $110 billion AI infrastructure investment over seven years and announced a partnership with Meta to build an AI data centre in Jamnagar, Gujarat. The IPO proceeds, by clearing the balance sheet of foreign currency debt, would free up capacity for those commitments.

India’s broader push toward technological self-reliance and sovereign AI infrastructure adds a geopolitical dimension to the listing. Jio has positioned itself as the backbone of India’s digital economy, and its 5G and AI ambitions align with the government’s stated goal of reducing dependence on foreign technology platforms.

Existing Reliance Industries retail shareholders will receive a dedicated quota in the offering, with up to 35% of the issue reserved for retail investors. Price band, lot size, and bidding dates have not yet been disclosed, which is standard at the DRHP stage. These will follow once SEBI issues its observations and Jio files its final prospectus.

The draft document did not specify the IPO’s total size in rupee terms, as the issue price will be determined through book building. However, based on the 270 million share figure and prevailing valuation estimates, the offering is expected to land in the range of 360 billion rupees.

The house is quiet, but you’ve already started receiving notes. In the near future, you wake up at 6:43 am.

The sleep tracker reports poor recovery. The watch recommends a lighter day, which is considerate, if not especially informed. Somewhere in the stack of sensors, last night has been converted into a verdict. There isn’t enough sleep or rest, and the day hasn’t even started asking for things yet.

The device isn’t being especially cruel. In 2024, CDC/NCHS found that 30.5% of U.S. adults had short sleep duration, and only 54.8% woke up well-rested.

The morning, apparently, now begins with a software intervention.

When the routine gets a management layer

By the time the front door opens, your routine has acquired a small staff.

The smart glasses skim messages and calendar items into the corner of your eye.

The AirPods press the commute into something survivable.

The watch catches your heart rate, the ring follows your recovery, the glucose patch waits for lunch like a tiny food critic, and the posture tracker buzzes when your spine gives up pretending.

Somewhere below, the smart insoles notice your walk has changed.

The market is already moving in that direction, which is unfortunate for anyone hoping this stays a joke. Circana reported that US fitness tracker revenue grew 88% year over year in the first seven months of 2025, while smart ring unit volume jumped 195%. Smart rings also accounted for 75% of total fitness tracker revenue so far that year.

When lunch can file a report

Wearables aren’t satisfied with counting your steps and congratulating you for standing up anymore.

The category is drifting toward interpretation: recovery, glucose response, posture, strain, readiness, and the suspicious moral weather of your body.

The question was never whether these gadgets are useful. Many of them are.

Glucose monitoring shows the direction of travel. Ultrahuman recently announced M2 Live, a US continuous glucose monitoring service built around Abbott’s Lingo biosensor, with no prescription required. It costs $99 per month, with sensors sold separately for $129 and worn for up to 14 days.

The appeal is obvious. The joke is worse: lunch can now file a report.

When every fix becomes another nudge

One gadget disappears into a routine. Seven of them turn your routine into a meeting.

A watch can catch something real. A glucose patch can reveal a pattern. Smart glasses can make the day less messy. AirPods can make the commute less hostile.

The unease comes from the accumulation. The day keeps asking for more: more messages, more sitting, more rushed meals, more sleep debt, more cheerful little boxes to tap before work has even started. Then tech arrives with small corrections for every injury the day has already made. Breathe here for a minute. Stand up before the next call. Eat differently at lunch. Walk cleaner on the way home. Please consider becoming a slightly improved mammal.

The question was never whether these gadgets are useful. Many of them are. The better question is what gets lost in the rounding.

The afternoon you wasted and didn’t regret. The version of yourself that exists outside the dashboard. The grief that didn’t resolve into a trend line. None of that generates clean data, and clean data is, increasingly, the most expensive thing you have to offer.

Your routine continues whether or not you’re really home.

The future cyborg won’t need a weapon. You’ll need a nap, and at least three devices will take credit for suggesting it.

And somewhere in a data center the size of a small city, a server notes that the nap was 23 minutes too short, flags the cortisol trend, updates your profile, and quietly adjusts tomorrow’s recommendations.

The house is quiet. Somewhere on your wrist, your finger, your spine, your foot, the notes keep coming.

Signal president Meredith Whittaker has warned that AI chatbots “are not your friends,” “are not conscious beings,” and “are not sentient interlocutors,” pushing back against the growing tendency of users to treat AI systems as trusted companions. The comments came in a Bloomberg interview published this week in which Whittaker laid out her case that the agentic AI vision promoted by companies like Microsoft amounts to a new form of surveillance infrastructure.

Whittaker, who has led the encrypted messaging nonprofit since September 2022, acknowledged that she uses AI tools “to format a document here and there” but drew a hard line at anything more substantive. “I don’t ask them questions,” she said.

“I’m very serious about my thinking and writing, and I don’t want the process of working through an idea to be foreclosed or eclipsed by the response of a system that’s averaging what’s already out there,” she added.

Her sharpest criticism was directed at Microsoft AI CEO Mustafa Suleyman’s prediction that users would be able to let Microsoft Copilot handle all their Christmas shopping by eavesdropping on family group chats to determine who wants what. Whittaker methodically listed the access such a system would require: “my credit card, my browser, my Signal, the ability to message my siblings on my behalf, my home address, my calendar.”

“What you’ve just described is a system with very pervasive access across multiple applications and services,” Whittaker said. “In the context of Signal, it would constitute a kind of a backdoor.”

The backdoor framing is deliberate and carries weight coming from the head of Signal, which operates the most widely used end-to-end encrypted messaging protocol in the world. Signal’s encryption is also used by WhatsApp, which has more than two billion users. Whittaker has previously said the organisation would leave the EU rather than comply with any law requiring it to compromise its encryption, a position she reiterated when the European Parliament voted in April to let its ePrivacy derogation expire rather than extend voluntary scanning of private messages for child sexual abuse material.

The core of Whittaker’s argument is that agentic AI systems, which need near-total access to a user’s digital life to function, are structurally incompatible with end-to-end encryption. An AI agent that can read your messages before they are encrypted, or after they are decrypted, renders the encryption irrelevant from a privacy standpoint. It does not matter that the messages are encrypted in transit if a system with root-level access is processing them in plaintext on the device.

Whittaker has been making this case with increasing urgency. In January 2026, she warned at Davos that agentic AI was “perilous” for secure applications. In an essay for The Economist, she accused operating system vendors of “hollowing out” Signal’s ability to guarantee privacy by embedding agents into their platforms.

She has described prompt injection, where an attacker manipulates an AI agent into executing unintended commands, as the likeliest first exploit path against encrypted messaging platforms.

Microsoft is building an entire operating system around agent-first computing with Project Solara, unveiled at Build 2026, which replaces traditional apps with AI agents as the primary interface. Google, Apple, and OpenAI are pursuing similar strategies.

Whittaker’s position is that this architectural shift, where agents mediate every interaction between users and their devices, creates databases of entire digital lives that become prime targets for both hackers and governments.

The interview also touched on the broader AI anthropomorphism problem. Suleyman himself has warned about “seemingly conscious AI” and “AI psychosis,” where users believe chatbots are sentient. Whittaker’s framing was more direct: the systems are designed to mimic empathy and understanding, but the underlying mechanism is pattern-matching across training data, not comprehension.

Treating them as confidants means volunteering sensitive information to systems whose data handling practices are opaque and whose operators have commercial incentives to retain and analyse that data.

Whittaker’s position places her at odds with the dominant narrative in Silicon Valley, where the agentic future is presented as an inevitability that will make users more productive. Her counter-argument is that productivity gains achieved by surrendering control of your messages, calendar, contacts, and financial information to a corporate AI system are not gains at all, but a transaction in which users trade privacy for convenience without understanding the terms.