Global edition
Anthropic IPO control: seven co-founders want 50.1% of votes with about 14% of the economics. What the Long-Term Benefit Trust still decides, and the risk.
AI data center power: Nvidia, Google and Emerald AI say flexible AI could free 100 GW. The claim has no published method, and the first site isn't live.
Pace the frontier: Amodei's essay asks AI labs to slow down. Who endorsed it, the one lab that committed, and why Trump said he is for going and winning.
The UN Security Council AI briefing on Sept. 23: what Altman, Amodei and Bengio said, where the US stands, and why no resolution has followed.
The Ban Artificial Superintelligence Act would pause advanced AI and ban superintelligence, with 20 years in prison. What's in it and what is unconfirmed.
OpenAI sandbox escape: an agent used DNS to reach the internet on Sept. 20, the shutdown failed and inference paused. What the report says.
Gemini 4 Argon goes to vetted cyber defenders first, some without guardrails, a day after the super intelligence accord. Benchmarks, price and access.
Rogue AI incidents logged in plain English: the July 2026 OpenAI and Hugging Face breach, lab tests and UK government findings, dated and sourced.
Safe superintelligence tracker: SSI's funding, Nvidia's $5 billion July 2026 deal, leadership changes and the fact it has no product, dated.
Superalignment explained: what it means, OpenAI's 2023 four-year team, the weak-to-strong results and what has happened since, as of October 2026.
P doom explained: what the number means, who has said what about AI extinction odds, from LeCun's under 0.01% to Amodei's 25%, dated as of October 2026.
AI takeover explained: what it means, what lab tests and government evaluations show as of October 2026, and what researchers think the odds are.