Plus: OpenAI has unveiled its long-awaited super app. The AI firm Anthropic has got the clearest glimpse yet at what's really going on inside large language models as they answer questions or carry out tasks. What they found ranges from the mundane to the unnerving. Researchers at the company built a tool called the Jacobian lens (or J-lens) and used it to uncover a hidden area, which they named the J-space, inside its flagship LLM, Claude. The J-space contains words related to the response a model is working on but may not ultimately produce. If Claude were a person (which it is not), you might say these hidden words reveal what's on its mind before it actually speaks.

Amid live coding sessions and Silicon Valley optimism, the UN's AI for Good summit wrestled with an increasingly urgent question: Can global governance catch up before the technology races beyond its control? Dodge past the live onstage coding sessions, AI refresher courses, an obstacle course of gizmos, round people walking round with glowing green silent-disco-style headphones blaring UN panel discussions into your ears, and you can take a pause for breath. But you might find yourself in the Networking Zone, on a rotating seating contraption called UFOTECH that looks more like the kind of lazy Susan you'd encounter at a Chinese restaurant than the networking bench it is designed to function as. This is the AI for Good summit, organized by the United Nations' International Telecommunication Union (ITU), where representatives from the private and public sectors try to discuss how to harness the technology for the benefit, rather than the detriment, of humanity. While Silicon Valley execs and AI lab leaders are testifying to lawmakers in Washington about the risks of superintelligence, and the White House slaps export controls on chips, the UN AI for Good Summit--now in its 10th year--is focused on much more idealistic goals.

The move comes after Simo took significant medical leave. She will stay on as a part-time adviser. OpenAI's chief executive of AGI deployment, Fidji Simo, is leaving her full-time role at the company and transitioning to become a part-time adviser. The move comes after Simo took a monthslong medical leave due to a worsening neuroimmune condition. "Three months ago, I had to go on medical leave after a severe exacerbation of a chronic illness I've lived with for seven years," Simo wrote in a post Thursday on X. "During that time, it became clear that the road to recovery would be much longer and more complex than I had anticipated--and that I needed to focus on it fully."

The AI firm Anthropic has developed a technique that has given it the clearest glimpse yet at what's really going on inside large language models as they answer questions or carry out tasks. What they found ranges from the mundane to the unnerving. Researchers at the company built a tool called the Jacobian lens (or J-lens) and used it to uncover a hidden area, which they named the J-space, inside Claude Opus 4.6, a version of Anthropic's flagship LLM released in February.

Silicon Valley is making a show of helping prepare the country for AI layoffs. In late March, I started receiving daily texts from the federal government about AI. " AI is changing how we work and live," one message read. "You might feel curious, skeptical, or unsure--that's normal." I had enrolled in an AI-literacy course from the Labor Department created to help workers succeed in the ChatGPT economy. The weeklong program, created in partnership with an AI start-up and delivered by text message, was supposed to equip Americans with "foundational AI skills," according to an agency press release.

OpenAI released its latest advanced AI model, called ChatGPT 5.6. OpenAI released its latest advanced AI model, called ChatGPT 5.6. Staggered release of ChatGPT 5.6 follows similar restrictions on rival firm Anthropic's latest AI models OpenAI released its latest advanced AI model, called ChatGPT 5.6, on Thursday after earlier delaying the public rollout over US government concerns about cybersecurity. The Trump administration had requested last month that OpenAI limit the release to a small group of government-approved users. OpenAI complied with the White House's request last month.

Claude subscribers must soon pay usage-based fees to access Anthropic's best consumer AI model--a sign that the golden era of AI subscriptions is ending. AI model developers have long offered consumers a simple deal: Use our technology for free through an online chatbot, or pay a monthly subscription to receive more usage, premium features, and advanced models. Anthropic is about to make that bargain a lot more complicated. Starting on July 12 at 11:59PM PT, subscribers to Anthropic's $20, $100, and $200-a-month plans will need to pay additional usage-based fees to access Claude Fable 5, the consumer version of the company's highly capable Mythos 5 AI model . This appears to be the first time a frontier AI lab has gated a consumer AI model behind usage-based billing.

Companies will once again be allowed to scan citizens' personal texts, emails, and social media messages via the "chat control" bill to find child abuse material online. The European Parliament has voted to extend legislation allowing tech companies to voluntarily scan users' private messages for child sexual abuse material, despite a majority of lawmakers voting against the proposal. The ruling reinstates permissions for firms including Meta, Google, and Microsoft to scan private text, email, and social media messages through a bill nicknamed "Chat Control" by critics. End-to-end encrypted chats, such as those on WhatsApp and Signal, remain exempt. "It will mean that private companies may deny your right to have confidential digital conversations," Simeon de Brouwer, policy advisor at Brussels-based advocacy group European Digital Rights tells WIRED, "they could, if they want to, read every message you write, every email you send, every picture you share."

Plus: NATO is building a network to stop Russian attackers in their tracks. I was really looking forward to July 4, and not just because I love a poolside barbecue. This year the American holiday also marked a big symbolic deadline for US nuclear power. Last year the Trump administration set a goal to see three new microreactors achieve criticality, a technical milestone establishing that a reactor can sustain a chain reaction, by the nation's 250th birthday. And just in time, not just three, but four reactors did so. But achieving criticality doesn't mean a reactor is ready to provide electricity for the grid (or at all, for that matter).

The $28 Million Mistake That Inspired Estonia's AI "Fuckup Finder" Now Estonia is using AI to spot legal errors before they become law--and to automate more of the state. Estonia's AI embarrassment began with a single wrong phrase. In December, the Riigikogu, Estonia's parliament, passed changes to the country's Gambling Tax Act meant to lower the tax rate on remote gambling. But the wording of the law referred only to "skill games" for that year, not games of chance or remote gambling. Estonia's entire gambling industry is worth around €300 million ($343 million), and its online gambling market is one of the fastest growing in the EU.