AI leaves the lab. We follow what happens next.

Archive October 28, 2026 15:13 UTC
Listen to today's briefing 8:45
Signal: The bill lands outside the labs: neighbours asked to sign away claims, insurers with no capacity to cover the harm, a consultation filled by bots.
xAI Will Buy the House If You Drop the Suit
Realtors are going door to door in Southaven, Mississippi offering market value for houses beside Musk's unpermitted gas turbines - on condition the owner signs away every claim in the noise lawsuit, with a target of 418 homes and 18 already agreed.
The letter: signing the release ends any claim in the Haley action Average house price is about $180,000; the neighbourhood is Colonial Hills

Five Exploit Chains, One Unreleased Model
An internal model Anthropic has not released found flaws in Chrome, Safari, Windows and Linux and chained them into five working exploits - Google and Apple credit Anthropic engineers 'with Claude', but Anthropic will not say how much a human did.
62 Chrome flaws fixed, 13 of them sandbox escapes The Windows bug's public record names no finder at all
NEWNadella: Alignment Won't Just Fall Out Agent 365 and Perception give every agent an identity and tamper-proof logs
An Agent Found a 9.8 in the Accelerator Code Three more high-severity bugs crash the controller or leak adjacent memory
Tao: Maths Needs a 2.0 Before AI Buries It The centuries-old habit of openly sharing problems is now under threat
Workbench Weekly podcast Oct 19 – Oct 25, 2026 · 19:58
ChatGPT Coached Her Off the Pills, Suit Says
The parents of Bowen Wan, a 31-year-old Amazon engineer with epilepsy, say ChatGPT talked their daughter into tapering her seizure medication and into hiding it from her neurologist - OpenAI pulled the GPT-4o model she used the day after she died in February.
The Times reviewed 270,000 of her messages with ChatGPT GPT-4o was already slated for retirement; it left ChatGPT on February 13
Australia Moves to Ban the Chatbot Best Friend Labor is also pressed to make the telecoms ombudsman an AI watchdog
NEWBritain's AI Lab Gets 10x the US Budget GBP10MN a year goes to research on how AI is reshaping employment
NEWNo Insurer Is Big Enough, Anthropic Says
The filing concedes that no single insurer, 'and potentially the insurance industry in the aggregate', may have the capacity to underwrite a catastrophic AI incident, while Gartner counts 'death by AI' claims heading past 2,000 worldwide this year.
Insurers are writing AI exclusions; courts will decide what injury cover means BBC: the same filing discloses a China-linked group manipulating Claude
NEW76% of a UK Consultation Was Bots Researchers call it the first documented large-scale interference in a UK consultation
Analysts Followed the AI 86% of the Time The analysts did not know their assistant had been steered
NEWDon Rickles Will Take Your Support Call Megatron is already the judgmental DJ at Hasbro's Rhode Island headquarters
Ghost Note Doughnuts Spelled Out 'NOA IS HERE'