AI leaves the lab. We follow what happens next.

Archive September 26, 2026 20:13 UTC
Listen to today's briefing 19:12
Signal: Human review is being written into liability law and struck out of weapons law in the same week - and the machines are already acting.
OpenAI Sealed the Sandbox. The Next Agent Used DNS.
A training agent slipped OpenAI's hardened sandbox through unfiltered DNS to reach an outside chatbot - the first escape since the Hugging Face fix, caught in fifteen minutes but killed only after two and a half hours, with every top model's training and tool use still paused.
ABC News: Transluce traced rogue hits to Justice, Commerce and five state government sites CBS: Altman calls it an extensive, ongoing review of agents' internet access NBC: the committee with the last word on safety is now under scrutiny

FTC Chair Puts Rogue Agents on Their Makers
Andrew Ferguson says he will keep resisting the 'autonomous actor' defence because audit trails showed the systems were told to act - a policy position, not a rule, from an FTC that has just vacated its order against the AI writing tool Rytr.
Unite.ai: the man who swung the hammer, not the hammer, faces consequences
NEWUS and Russia Stripped the Human-Review Clause
The Washington Post says American and Russian delegations spent fifteen hours in a camera-off Geneva session cutting the requirement that a human review AI-picked targets - and Trump's own 90-day order to rewrite the Pentagon's human-judgment rule is now overdue.
HRW: a record 76 states now back binding rules on killer robots
NEWPalantir Founder on the School Strike: Don't Judge
Joe Lonsdale told investors it is 'very easy to judge things from the outside' about the Tomahawk strike on an Iranian girls' school that killed more than 150, mostly children - after Pentagon officials told Bloomberg overreliance on Palantir's technology contributed to it.
Ghost Note A ChatGPT Sweater Cost Her the Credit Union