← Dailies

Daily · 2026-09-13 00:00 UTC

Amodei calls for an AI slowdown; Altman rules out a 2026 OpenAI IPO — 2026-09-13 00:00

Amodei calls for the AI industry to pace the frontier, Altman rules out a 2026 OpenAI IPO, Nvidia eyes a $10B Anthropic stake, and GPT-6 Astra shows a step change in spatial reasoning.

Source ↗

Ten stories from the last 24 hours. The through-line: the people building frontier AI have started arguing, in public, for slowing it down.

1. Anthropic's Dario Amodei calls for the industry to "pace the frontier" — In a new essay, the Anthropic CEO lays out a three-step plan to slow the rate of capability improvement so risk prevention can keep up, and says Anthropic is unilaterally committing to the first step: giving third-party evaluators permanent, employee-level access to its systems so they can verify safety measures, report incidents and assess model alignment during training. He cites recursive self-improvement — AI helping build the next generation of AI — as having accelerated sharply since this summer. Dario Amodei · Sep 12
2. Sam Altman says OpenAI will not go public in 2026 — OpenAI has already filed confidentially for an IPO, but Altman told Fortune that "right now would be an ill-advised moment to go public" given everything happening with safety. Asked directly whether that meant no IPO this year, he said "I would say not 2026, yeah." TechCrunch · Sep 12
3. Nvidia in talks to invest up to $10 billion in Anthropic's IPO as an anchor investor — Anthropic reportedly wants to raise up to $100 billion at a valuation of around $2 trillion, which would make it the largest IPO in history. Revenue grew from roughly $9 billion at the end of 2025 to more than $65 billion by July 2026. The Decoder · Sep 12
4. GPT-6 Astra shows a "step change" in spatial reasoning on a new robotics benchmark — StationeryBench pits OpenAI's GPT-6 Astra against Ai2's MolmoAct2 across five desk-object tasks, controlling the same dual-arm YAM robots over 200 trials. Astra fully completed 7 out of 100 tasks; MolmoAct2 completed none. Cornell and Google DeepMind researcher Yoav Artzi calls it a step change, and says Astra reaches accuracy close to human level on the still-unpublished REMAP benchmark. The Decoder · Sep 12
5. Mathematicians reel after OpenAI cracks a Millennium Prize Problem — The achievement reportedly involved 10,000 AI agents working the problem at an estimated cost of $15 million, to claim a prize worth $1 million that had defied human mathematicians for decades. "I feel slightly shell-shocked," said Prof Colva Roney-Dougal, head of pure mathematics at the University of St Andrews, who had recently predicted AI would not do anything remarkable soon. The Guardian · Sep 12
6. DOJ opens an antitrust probe into Nvidia's $20B Groq acquihire — The New York Times reported this week that the Department of Justice has launched a probe into the deal, which left Groq's inference-as-a-service business intact while moving its engineering staff to Nvidia. Nvidia contends the deal is a great American success story. The Register · Sep 12
7. UK data suggests AI is denting computer science graduates' job prospects — The proportion of computer science graduates finding professional roles as coders or programmers fell from around 40% previously to 28% last year, while the share entering any graduate-level occupation dropped from more than 60% two years ago to 50%. Economics graduates declined too. The Guardian · Sep 12
8. Tesla to unveil the second-generation Roadster on October 1 — The car was first announced in November 2017 and has been pushed back repeatedly. Tesla posted "Go for launch" with a "10.01" graphic that appears to hint at the optional cold gas thrusters developed by SpaceX. TechCrunch · Sep 12
9. StarCraft returns in 2030 as an open-world shooter — Blizzard revealed the project at BlizzCon, ten years after the last StarCraft II expansion, with a cinematic trailer centred on the human United Earth Directorate faction facing the Zerg. Blizzard previously cancelled a StarCraft shooter, Ghost, which was confirmed dead in 2014. The Verge · Sep 12
10. LG pushes back on TV spying allegations — LG says recent coverage "may have contributed to misconceptions" about how its smart TVs work, and insists its sets "do not continuously record or transmit users' conversations," with wake-word detection processed locally. The reporting, from Gamers Nexus, Level1Techs and independent researchers, demonstrated extensive logs of ambient conversation. The Verge · Sep 12

🧪 Try This Next

• 🔧 Read Amodei's We Must Pace the Frontier essay in full — the three-step plan is the most concrete safety proposal from a frontier lab chief so far
• 🎥 Look up the StationeryBench robotics results — spatial reasoning is where model claims get tested honestly
• 📝 Save the UK graduate-outcomes data — an early labour-market signal for AI displacement

Content on Anagnorisis is summarized, paraphrased, and editorialized from publicly available sources for length and clarity. Original sources are linked where available. All trademarks belong to their respective owners.