Whitehall has spent time and taxpayer money confirming that teenagers who spend less time on social media tend to be happier, better rested, and more willing to interact with the people sitting in the same house. Ministers say the government-backed trial strengthens the case for their planned ban on social media access for under-16s after providing real-world evidence that cutting back on apps does exactly what parents have been saying for years. The study followed more than 300 families across the UK as they tried one of three restrictions at home: removing social media apps entirely, blocking access between 9 pm. and 7 am, or limiting each app to 15 minutes a day. Across the trial, teenagers reported going to bed earlier, sleeping better, feeling calmer, and concentrating more effectively at school. Parents also described quieter evenings and more time spent talking together instead of staring at separate screens. The strongest effects came when social media apps disappeared altogether. Those families reported the biggest increases in face-to-face time with friends and evenings spent together, alongside less screen time and better focus. Families trying the overnight curfew also saw benefits, particularly around sleep, with some parents saying the 9 pm cutoff became routine within a couple of weeks. The 15-minute limit, by contrast, flopped. Teenagers found it too restrictive to use apps in any meaningful way, leading many to ration their allotted minutes, switch to other devices, or simply migrate to platforms that weren’t subject to the same restrictions. Secretary of State Liz Kendall said the findings reflected what many parents already believed: “These findings show what parents have been telling us all along: when children spend less time on social media, the benefits are real.” “It’s why we’re taking the strongest action in the world to support a generation that is healthier, happier and more connected to the people and experiences that matter most – not just to their screens,” she added. The announcement also attempts to head off one of the biggest criticisms of the government’s forthcoming age restrictions: that teenagers will simply install a VPN and carry on regardless. New figures released alongside the study suggest that while roughly a quarter of children aged 11 to 17 have used a VPN, only between 7 percent and 10 percent say they do so specifically to bypass age checks. Instead, the government says the far more common workaround is simply entering a fake date of birth, a trick it argues will become less effective once platforms are required to use stronger age-verification systems. None of this proves the government’s under-16 social media ban will be painless to enforce, or universally popular. But if ministers were looking for evidence that fewer hours of doomscrolling might leave teenagers a little better rested and families a little less distracted, they now have a government-funded study telling them exactly that. ®

Claude looks substantially more token-hungry than OpenAI’s GPT-5.x, thanks to the new tokenizer that Anthropic shipped with recent releases. Large language models (LLMs) use tokenizers to handle the mapping of text into tokens. There’s no set definition of a token, but they’re typically a set of three or four characters that are mapped to the integers LLMs actually process. Tokens have become the basic economic unit for billing use of AI models. Because the slicing of words into tokens and the tokens required per task vary across models, it has become rather difficult to predict the final bill for playing the AI slot machine. Recent changes to Anthropic’s tokenizer appear to have further complicated matters by making the same content more costly to process on certain models. Playcode, an AI app building platform, recently analyzed the impact of Anthropic’s latest tokenizer and found that the same TypeScript file processed by Claude can consume up to 73 percent more tokens than OpenAI’s GPT-5.x model family. Anthropic acknowledges its new tokenizer – announced at the end of June when Sonnet 5 shipped – may generate more tokens for the same input than prior versions. “Sonnet 5 is an upgrade to Sonnet 4.6, but it uses an updated tokenizer that changes how the model processes text to improve performance (this is similar to the tokenizer change we introduced with Claude Opus 4.7),” the company explained. “The tradeoff is that the same input can map to more tokens: roughly 1.0–1.35× depending on the content type.” Anthropic offered Sonnet at a reduced introductory rate – $2/million input tokens and $10/million output tokens through August 31, 2026 – to make the inflated token generation more or less cost-neutral. But the price is set to rise to $3/million and $15/millionafter that. Anthropic did not immediately respond to a request for comment. The company says that users of its new tokenizer may see their bills rise by as much as a third compared to the tokenizer it used with its older models. Costs for Anthropic users could go even higher when compared against the latest iteration of OpenAI’s o200k tokenizer. Playcode’s cross-vendor token comparison finds that for a 2,888 character TypeScript file, Claude’s new tokenizer emits 1.73x more tokens than GPT-5.x’s tokenizer and 1.32x more than Claude’s old tokenizer. These figures differ for different types of code: Rust taps in at 1.58x, JavaScript 1.52x, and Python 1.50x. So if Anthropic’s list prices were adjusted to be comparable with OpenAI’s GPT-5.x baseline, Playcode suggests Opus 4.8’s cost would be $7.50/M input and $37.50/M output instead of the published figure of $5/M and $25/M. The AI app platform notes that a team from marketing platform Ploy this week published an account of a production migration using OpenAI’s GPT-5.6 Sol and Anthropic’s Opus 4.8. “GPT-5.6 finished pages 2.2× faster, cost 27 percent less, and used about half the output tokens,” Ploy claimed. There are other factors that go into calculating AI bills. As we noted recently, costs should be judged in terms of task completion and the impact of model harnesses (e.g. Claude Code, Codex, Pi, OpenCode, etc.) should also be evaluated when attempting to calculate the cost of running AI workloads. ®

IBM has again teased small hardware, this time in the form of an update for its smallest POWER server. The model S1112, teased Tuesday in a customer announcement, is a 2U, single-socket POWER11 server IBM offers in rack-mountable and what the company calls “Tower/deskside configuration.” The rackable model can handle a ten-core POWER processor. The Tower/deskside form factor machine must make do with a four-core engine IBM seems to have two roles in mind for the new machines: edge deployments and standalone use by those who are taking their first strides into using the last remaining proprietary minicomputer ecosystem. One is edge deployments. The other is as an entry-level box, with the description of the tower unit suggesting its very existence means “even the smallest customers” can use it as an on-ramp to more POWER implementations. News of the S1112 marks the second time in a week that IBM has gone low with modest hardware. Last week Big Blue teased the z17 ME2, a rackable mainframe that it said completed its range by offering a smaller and cheaper piece of hardware. The twin launches continue IBM’s policy of creating smaller versions of its enterprise hardware, albeit well after launch of big iron: the first POWER 11 boxes landed in July 2025 and the first z17 series mainframes debuted in April of the same year. The S1112 includes a quartet of DIMM slots and can handle up to 512GB of DDR5 memory. The box runs IBM i, AIX, and Linux – or all three because it supports IBM’s PowerVM virtualization tools. In an almost certain non-coincidence, Big Blue on Tuesday also announced upgrades for PowerVM including improved automation and support for the S1112. Big Blue has also looked after users of bigger POWER fleets, by expanding the number of Spyre accelerators – IBM’s neural processing units – that POWER servers can support from eight to twelve. IBM pitches POWER as a capable AI platform, so allowing it to use more accelerators can’t hurt. IBM plans to start selling most of the kit described above on July 24, although customers who crave the S1112 to deploy in Taiwan will have to wait until September. 15. Would-be buyers in South Africa, India, and China must wait longer still, until December 11. ®

Indian tech services giant and retro software house HCL has decided to get into the AI datacenter business. The company yesterday revealed its plan in an announcement [PDF] released alongside its Q1 results, which included news of three-percent year-over-year revenue growth to $3.65 billion and 20 percent growth in net income which reached $488 million. CEO C. Vijayakumar also pointed to 62 percent year-over-year revenue growth for a segment HCL calls “Advanced AI” that encompasses building its own AI platforms. The CEO said HCL’s strategy is to “Benefit disproportionately from the AI-native and AI-amplified opportunities” because they “together represent the fastest growing pool of enterprise spend.” The company has therefore decided to get into the datacenter business and has found ₹3,500 crore ($36.5 million) to put toward facilities it says have “potential to scale to 50MW of capacity.” That’s not a vast facility – just one of Meta’s datacenters will host 50GW of kit – but Vijayakumar said HCL can make it relevant by using its existing software to offer “full-stack” infrastructure. “The biggest opportunity is not to rent AI, but to own the full stack,” the CEO said. “The datacenters that compute the models built to address client-specific needs.” “This is a business which is shifting from physical infrastructure to higher value AI-ready solutions,” he added. “We will create full-stack offerings by combining our capabilities across AI datacenter design, DevOps, and cloud operations, as well as a software portfolio with our new datacenter business.” HCL’s focus appears to be on Indian customers, as Vijayakumar said the datacenter investment will “position us as a key enabler of India’s sovereign AI ecosystem, expanding our presence in the fastest-growing market among largest economies with differentiated offerings around sovereign cloud, secure AI, and managed AI infrastructure.” The CEO said HCL is already “in advanced discussions with clients to ensure we start with certain level of committed consumption from day one.” The company didn’t say where it will build its bit barns, when they might come online, or how it will secure energy supply – an important consideration given we yesterday reported on an effort to locate a datacenter in renewable-energy-rich Bhutan to serve Indian customers. Vijayakumar also revealed that HCL booked $2.4 billion of new business in the quarter, a record. The CEO pointed to one of those deals as an exemplar of HCL’s AI smarts, as it will see the services company work with an unnamed Fortune 250 semiconductor equipment OEM “to accelerate AI-driven transformation across its semiconductor engineering and manufacturing value stream.” To make that happen, HCL will deploy SAP, integrate it with existing systems, and establish “an enterprise backbone for a future-ready, scalable, AI-led digital supply chain.” Another new deal, struck earlier this month and therefore not included in the $2.4 billion of new deals won in the quarter ended June 30, will see HCL work with an unidentified “Europe-headquartered Fortune Global 50 firm as a technology partner to accelerate AI-led transformation and management of their digital workplace and enterprise networks.” Numerous reports in Indian media identified the new client as Mercedes Benz, and suggest the automotive giant has moved its business to HCL from Infosys, which announces its quarterly results next week. ®

The hardest problem in AI is no longer the chip but the megawatt. For much of the past three years, the global AI race has focused on semiconductors, with governments competing for advanced chips, technology outfits scrambling to secure GPUs, and investors pouring billions into ever larger datacenters. Yet the binding constraint has shifted from compute to the power required to run it. For anyone trying to energize a new AI cluster today, the bottleneck is rarely silicon; it is grid access, interconnection delays, and aging infrastructure. That was the central message from Envision founder and CEO Lei Zhang at VivaTech in Paris this June, where he argued that AI amounts to an energy revolution as much as a computing one. The steam engine transformed the industrial age by converting coal into motion, and the GPU now transforms the AI age by converting electricity into intelligence. History offers another lesson: James Watt changed industry through the efficient use of energy rather than by producing more steam. AI faces the same problem today, because the binding constraint has shifted from how many chips can be built to how they can be powered. The real risk: AI competing with society for energy The numbers behind the argument are stark. Goldman puts US datacenter power demand at 31 GW in 2025, rising to 66 GW by 2027, while assuming only about 72 percent of scheduled facilities arrive on time because electricity, not construction, is what typically slips. The IEA estimates that datacenters consumed roughly 1.5 percent of world electricity in 2024, a share rising to 3 percent by 2030 as AI-specific demand triples. The structural mismatch sits at the heart of the problem: AI models iterate every six months and chips refresh annually, while power grids have changed little in decades. Rack densities that sat at 5 kW are climbing toward 200 kW, and the IEA notes that AI server power density rose elevenfold between 2020 and 2025, with a further fourfold rise expected by 2027, straining the supply chains for power electronics and transformers that keep a cluster stable. The growing gap raises broader questions about where the energy will come from and who will bear the cost. Around the world, communities are asking whether AI infrastructure should draw on electricity that households, factories, hospitals, and public services also depend upon, with familiar concerns surfacing about consumer bills, manufacturer access to limited grid capacity, and the burden that ever-larger models place on public infrastructure. Those questions have moved beyond the purely technical into the societal, because the future of AI cannot rest on a model in which humanity competes with AI for power. Mission Gobi: Let AI follow energy Envision’s answer, Mission Gobi, unveiled at VivaTech, aims to develop 5 GW of green AI computing capacity across deserts and arid regions by 2030. For decades energy followed computing, and Mission Gobi reverses that logic on the premise that in the AI era, computing may need to follow energy. The logic is grounded in geography, because deserts offer some of the world’s richest solar and wind resources alongside vast expanses of low-cost land, with the additional advantage of little competing residential or industrial demand. Rather than drawing power from homes, factories, and public services, Mission Gobi seeks to build entirely new renewable energy systems dedicated to AI, expanding the available supply instead of asking society to share a fixed pie. The philosophy reduces to a single idea: compute should chase power, not the other way around. The economics matter because electricity determines whether a facility is viable, with power consistently accounting for the single largest operating cost at a datacenter and some estimates placing it at as much as 60 percent of the operational budget. Building energy-native AI infrastructure Envision splits the system into three layers: an intelligent operating hub, Physical AI powered by its Tianji Weather Foundation Model and Dubhe Energy Foundation Model, and advanced power infrastructure. Together they integrate generation, storage, grid, power electronics, computing, and large-scale AI models into a unified architecture. The challenge lies in coordinating renewable power rather than merely generating it, because AI facilities require stable, high-quality electricity while solar and wind output fluctuate continuously. Envision argues that large-scale predictive models can help balance generation, storage, and demand in real time. The concept has already moved beyond theory. In Chifeng, Inner Mongolia, Envision runs a 2 GW system on 100 percent renewable energy, coordinating wind, solar, storage, hydrogen, and compute in real time, while a gigawatt-scale AI and computing campus in Ulanqab is being developed as a demonstration of what energy-native computing infrastructure could look like. A 5 GW pledge is ambitious, but the underlying read is sound: retrofitting decades-old city grids for gigawatt AI loads is a difficult undertaking, and purpose-built renewable compute, sited where power is cheapest, offers a credible alternative. SpaceX looks up, Mission Gobi looks out Envision is not alone in recognizing energy as AI’s defining constraint. Elon Musk’s SpaceX has explored concepts for orbital datacenters powered by uninterrupted solar energy in space, and the vision rests on the same recognition: the future bottleneck of AI may lie in energy rather than silicon. Both approaches seek to place computing where energy is most abundant. The two visions diverge in geography, with one reaching upward beyond Earth’s atmosphere and the other outward toward deserts and Gobi regions, though both start from the same premise: AI should not compete with humanity for power. A new blueprint for AI infrastructure If the industrial age was built around coal and the electrical age around power grids, the AI age may be built around energy abundance. The success of future AI infrastructure will not be measured by GPU counts and model sizes alone. It will also depend on whether the industry creates new energy supply, eases pressure on communities, and enables technological progress without reducing others’ access to power. Whether deserts become the preferred destination for future computing remains to be seen. What is becoming clear is that the next phase of the AI race will be defined not only by who builds the most powerful models, but by who can build the energy systems capable of sustaining them. The path forward runs through creating new energy supply rather than reallocating existing capacity away from households, factories, and public services. Contributed by Envision.

The federal government is warning users of home and small office routers to secure their devices as Russia state hackers continue to mass-compromise them for use in obscuring nefarious actions against sensitive organizations in the public and private sectors.

Both the Russian and Chinese governments have been compromising routers for years, sometimes in prolonged tugs-of-war to wrest control of devices the other has already commandeered. The US government has occasionally issued covert commands and taken other steps to disinfect routers. Google and other companies have also worked to disrupt the massive botnets that control compromised routers in lockstep. The actions to date are little more than whack-a-mole exercises as the operators simply replace their botnets with new ones.

Proxy networks: The go-to tool

“Russian Federal Security Service (FSB) Center 16 cyber actors continue to exploit poorly configured and vulnerable networking devices worldwide, opportunistically compromising multiple critical infrastructure sector networks,” the Cybersecurity and Infrastructure Security Agency said Monday. The hacking groups are tracked under various names, including Berserk Bear, Energetic Bear, Crouching Yeti, Dragonfly, Ghost Blizzard, and Static Tundra. The advisory was co-issued by governments from around the world, including Australia, Denmark, New Zealand, and the UK.

Read full article

Comments

Prompt injections, the malicious commands attackers embed into content to entice large language models to follow them, have been attackers’ go-to tool for turning AI platforms against their users. A well-phrased command sneaked into an email or calendar invitation is often all it takes to cause the LLM to exfiltrate sensitive data or follow other harmful actions.

Now, defenders are embracing the prompt injection, too.

A strong, sharp effect

Researchers from Tracebit on Monday said they found that placing prompt injections alongside passwords, cryptographic keys, and other secrets stored on Amazon Web Services was often all that was needed to shut down attacks from AI hacking agents. The prompts direct the attacking LLM to perform an action forbidden by its guardrails, the safety barriers AI developers erect to prevent it from taking harmful actions. The LLM responds by shutting down.

Read full article

Comments

A patch Microsoft released on Wednesday to fix a zero-day vulnerability in its Defender security engine may cause Windows machines to write files large enough to completely consume available disk space, the researcher who discovered the flaw said.

RoguePlanet, tracked as CVE-2026-50656, came to public notice in June when NightmareEclipse, the pseudonymous name used by a researcher, disclosed it along with code for exploiting it. The vulnerability allows remote attackers to gain administrative control of Windows 10 and Windows 11 machines, even when real-time protection has been disabled. Over the past few months, the anonymous researcher has published a handful of other zero-days that have sent Microsoft scrambling to develop patches.

Writing files of unlimited size

Microsoft said Wednesday that it patched RoguePlanet with an update to the Microsoft Malware Protection Engine, which is used by the Defender antivirus app. The fix will automatically be downloaded and installed without users having to take any action. Wednesday’s update also includes “defense-in-depth updates to help improve security-related features.”

Read full article

Comments

Allstate Insurance Company has accused Broadcom of haphazardly issuing audits against it because the insurance firm decided not to renew its contracts with VMware and CA Technologies.

The allegations were made in relation to a lawsuit that VMware filed against Allstate in December 2025, according to The Register. In the complaint, Broadcom alleges that Allstate failed to comply with license audits, which Broadcom claims its contract with Allstate requires.

In a June 12 filing, Allstate suggested that Broadcom issued the audits in response to Allstate deciding to end business with its companies. Allstate’s statement reads:

Read full article

Comments

A Linux vulnerability that allows untrusted virtual machines to gain root access to host machines is one of two high-severity flaws to surface this week in the open source operating system.

The vulnerability resides in KVM, which is, in essence, a virtual machine app included in the kernel of many Linux distributions. The vulnerability, tracked as CVE-2026-53359, allows guest virtual machines—such as those used in cloud platforms to isolate one user’s instance from the host OS and other user instances—to break out of that container.

Januscape: A threat to cloud platforms

The vulnerability affects KVM running on both AMD and Intel processors. It exploits bugs residing in the KVM guest-side, the portion of the VM that consists of only resources like the OS or drivers present in the guest VM, rather than resources present on the host machine. The threat went unnoticed in the Linux kernel for 16 years.

Read full article

Comments