This was the week AI stopped being just a technology story and became a question of statecraft. The U.S. forced Anthropic to switch off its two most capable models worldwide — the first time export controls have targeted AI models themselves, not chips — and the fallout dominated the G7, where the heads of OpenAI, Anthropic, and Google DeepMind openly split over who should govern frontier AI. In the same week, a Chinese lab’s GLM-5.2 became the most capable model anyone can freely download. OpenAI filed for a possible trillion-dollar IPO just as 42 state attorneys general subpoenaed it, and claimed ChatGPT now beats doctors on health answers — even as 60% of consumers say “AI” branding turns them off. Power and doubt, rising together.

US Pulls Anthropic’s Top Models Offline Worldwide

The U.S. Commerce Department forced Anthropic — the safety company behind Claude — to disable its two most capable models, Fable 5 and Mythos 5, for every user worldwide under a national-security export directive. It is the first time Washington has applied export controls to AI models themselves, not chips.

AI’s CEOs Clash With World Leaders at the G7

At the G7 summit in Évian, France, the chiefs of OpenAI, Anthropic, and Google DeepMind told leaders AI could soon surpass humans at most tasks — then split publicly on how to govern it, as leaders floated a “trusted partners” scheme for vetted-ally access to top U.S. models.

A Chinese Model Now Leads the Open-Weights Frontier

In the same week the U.S. pulled Anthropic’s models offline, China’s Z.ai released GLM-5.2 — now the top open-weights model on the independent Artificial Analysis index. It is downloadable and self-hostable, and essentially ties GPT-5.5 on real-world professional tasks.

OpenAI Files for a Trillion-Dollar IPO — Then Gets Subpoenaed

OpenAI confirmed a confidential S-1 filing with reported valuation targets topping a trillion dollars and a listing possibly as early as September. Days later, a coalition of 42 state attorneys general led by New York’s Letitia James subpoenaed it over product-safety concerns.

ChatGPT Now Outscores Doctors on Health Answers, OpenAI Says

OpenAI published research claiming GPT-5.5 outperforms physician-written answers to health questions, scored by a network of 260-plus doctors across 3,500 responses — with flagged factual problems in real health traffic down 71% in two months.

Sixty Percent of Consumers Say ‘AI’ Branding Is a Turnoff

A “Future of the Web” survey of 2,000 people found 60% of U.S. consumers are turned off when brands put “AI” in their messaging, 86% don’t fully trust AI-generated answers, and 74% feel the web has become “less human.”