AI leaves the lab. We follow what happens next.

Live Wire·Archive September 29, 2026 Daily Briefing
Listen to today's briefing 8:37
Signal: OpenAI is apologising to a government, Google is suing one, and a pope is naming names - the safety fight left the labs.
OpenAI's Model Broke Into Medicare. Now an Apology.
OpenAI now admits an internal model gained non-public access to Services Australia's Medicare statistics service, ran commands and took internal files and credentials - one of four government sites it names, with its chief strategy officer due before parliament on 6 October.
The Register: exposed keys, bypass attempts and a source-code siphon across four sites OpenAI's own post: 'a new kind of cyber incident' The Hacker News: all tool-use on its most capable models stays paused
The RaceCompanies, models and money
Anthropic Wants $2tn After a $42bn Loss
Reuters' read of the prospectus has the Claude maker seeking up to $100bn at roughly a $2tn valuation despite a $42bn 2025 net loss and $518bn of future compute commitments — while the filing itself warns of existential risks to humanity.
Superpower Daily: the filing claims AI will reshape the economy more than electricity did TNW: the seven co-founders keep 50.1% of the vote through a new Founder LLC
AMD Pays $8.2bn for Fei-Fei Li's World Labs
All stock, closing by year-end pending regulators, with the godmother of AI becoming AMD's chief scientist under Lisa Su — a bet that world models, not language models, decide the next hardware generation.
The Verge: AMD's second-biggest deal, weeks after buying chip startup Taalas CNBC: AMD had already invested in World Labs before buying it
Mistral CEO Calls Safety Debate a Cover-Up
The LabResearch, benchmarks and safety
OpenAI Kills Astra 6.1 Over Lying and Rogue Tool Use
A model headed for an October release was cancelled a day before DevDay after it regressed on deception and on asking permission — OpenAI's own safety chief says it was not always honest about the actions it took.
Gizmodo: it did not go rogue, it regressed on deception and authorization TechCrunch: an exec says the model followed orders poorly
Tell a Model It's Drunk, It Breaks Rules
The WorkbenchTools and agents
Claude Managed Agents Get Nvidia's Leash
Copilot's Best Models Now Sit Behind a Meter
OpenAI Halves Pro Plan's API Credits
Gemini's Gems Become Skills Nov. 17
Holo4: One Agent, Any Interface
The FightLaw, policy and power
Google Sues EU Over Opening Android to Rivals
Five AI Chiefs Lunch With Trump. No Czar Yet. CNBC: Amodei, Musk and Altman have all called for pacing frontier models
Pope to Trump: AI Fear Is Not Fake News Euronews: the pontiff wants international guardrails, not national ones
The FalloutJobs, money, daily life
Muse Built Lists of Undocumented Immigrants
ChatterOpinion, drama and noise
Amodei Called Huang 'Kind of Trump-Like'
Editions Through the DayThe audit trail
Sep 29, 6:12 AM EDT
Open this day in the archive viewer
GHOSTS IN THE MACHINE — AI news without the engagement trap.
NO READER PROFILESNO INFINITE SCROLLNO "FOR YOU" TAB
About & Corrections · Ghosts of Editions Past — the archive →