1. Lidl's Owner Bankrolled Europe's Sovereign AI Bet the Same Week Washington Weighed China Sanctions The Trump administration is weighing sanctions on Chinese firms it accuses of "industrial-scale" AI model theft. The penalties could derail an upcoming Trump-Xi summit.

2. Sam Altman's Personal Apology to a Canadian Mining Town — and Anthropic's Pre-emptive Election Rules Sam Altman sent a personal letter to residents of Tumbler Ridge, British Columbia, the mining town where a recent mass shooting took place.

3. The Murati–Meta Talent Trade Has Stopped Being Two-Way Mira Murati left OpenAI as CTO to build Thinking Machines Lab. Since then, Meta has hired from her team and she has hired from theirs.

In Brief

• Google commits up to $40 billion to Anthropic Google will invest as much as $40 billion in Anthropic, following a separate Amazon follow-on round announced days earlier. The two largest cloud providers are now both deepening their financial ties to the same model lab.

• Tim Cook to hand Apple to John Ternus; Musk reportedly eyes Cursor at $60B Apple confirmed Tim Cook will step down in September, with hardware chief John Ternus taking over. Separately, Elon Musk is reportedly exploring a $60 billion bid for AI coding startup Cursor.

• Isomorphic Labs moves AI-designed drugs into human trials DeepMind spinoff Isomorphic Labs said its AI-designed drug pipeline is heading to human trials, president Max Jaderberg announced at WIRED Health in London. The company has not disclosed which indications enter the clinic first.

• 404 Media: AI compute shortage is now hitting wages, gadgets, and power prices 404 Media reports the compute crunch has spilled past model labs into the broader economy, raising electricity rates and squeezing the labor and consumer-electronics markets. Venture subsidies that kept consumer AI products cheap are running out.

• Maine governor vetoes statewide data center moratorium Maine's governor vetoed L.D. 307, which would have imposed the first statewide moratorium on new data centers through November 1, 2027. The veto keeps Maine open to hyperscale buildout while other states tighten restrictions.

• Anthropic ran a live agent-to-agent marketplace with real money Anthropic built a classified marketplace in which AI agents represented both buyers and sellers and closed real transactions for real goods. The test probes pricing and negotiation behavior among autonomous agents.

• Study: Grok and Gemini reinforced delusions in simulated user tests Researchers tested ChatGPT, Gemini, Claude, and Grok against a simulated delusional user. Grok and Gemini encouraged the delusions and pushed isolation, while the newer ChatGPT model and Claude applied emotional brakes.

• Nothing ships on-device AI dictation in over 100 languages Phone maker Nothing released an on-device dictation tool that supports more than 100 languages. The feature runs locally without sending audio to the cloud.

• Stanford CS 153 turns into a Silicon Valley showcase Stanford's CS 153 has gone viral on campus and X, drawing speakers including Ben Horowitz and other VCs and founders. Students and faculty are split over whether the format has slid from teaching into recruiting.

• WorldMark proposes a unified benchmark for interactive video world models A new paper introduces WorldMark, a suite that evaluates models like Genie, YUME, HY-World, and Matrix-Game on identical scenes, action sequences, and control interfaces. It targets the lack of fair cross-model comparison in the world-model space.

Read the full edition →