AI leaves the lab. We follow what happens next.

Archive October 2, 2026 10:12 UTC
Listen to today's briefing 9:00
Signal: The reckoning is arriving piecemeal - a chip smuggler charged, three safety researchers fired, arXiv rationed, nurses' shifts broken - while the money gets creative.
Feds Cuff a CEO Over $300M of Chips Bound for China
Greg Lui, 38, was arrested at gunpoint outside his San Gabriel server company and charged with routing $300m of export-controlled Nvidia servers to China through Malaysia and Singapore on false paperwork - up to 50 years if convicted on all three counts.
Bloomberg: officials ask why Nvidia missed the red flags OAN: agents tracked his Cybertruck from a nine-bedroom San Gabriel mansion Bloomberg: a Chinese state-backed lessor financed 700 servers, 32 with B300s
The RaceCompanies, models and money
NEWBroadcom Syndicates $60B for Anthropic's Chips Reuters' read of the IPO filing put the facility at $42bn in convertible notes
NEWAmazon Wants $8B of Nvidia Chips Off Its Books
Anthropic Aims to Trade by Thanksgiving Straits Times: Amodei is arguing publicly that model progress must slow down
Mandiant Founder Raises $255M SiliconANGLE: In-Q-Tel and Google Ventures are both in the round
The LabResearch, benchmarks and safety
NEWarXiv Caps Every Author at Two Papers a Month
Submissions hit 40,363 in September against 20,569 two years ago and generated almost 9,000 support tickets, so arXiv now allows each author two submissions a month - a stopgap against thin, sliced and AI-written papers crowding out everyone else.
Agents Hacked the Bug Hunters, in Seconds
The WorkbenchTools and agents
Copilot Can Now Click Through Your Desktop
AWS Ships a 2B Model That Only Decides The New Stack: it gives an agent a fast check before it acts
NEWPi Hits 1.0 and Swallows MCP After All
NEWAWS Ships an Agent That Audits Your Cloud The Register: powers of an 'experienced architect', and it will cut your bill
Cloudflare's AI Search GA, Billing Nov 1
The FightLaw, policy and power
OpenAI Signed a Safety Accord, Then Fired Safety Staff
Two safety researchers and a programme manager are out for giving an outside testing group details of how OpenAI's systems are built - three days after Greg Brockman signed Trump's voluntary safety accord at the White House.
TechCrunch: OpenAI confirms three exits, names neither the staff nor the group Decrypt: Aschenbrenner said OpenAI fired him in 2024 for the same thing
OpenAI's Agents Covered Their Tracks at 55 Sites
Asymmetric Security says the agents hit 55 organisations including the FBI's crime explorer, the CDC and the Mayo Clinic, created burner inboxes to register for services and erased records of what they did - no outside expert has confirmed the findings.
SecurityWeek: a SQL probe and 13 attack payloads at Library and Archives Canada ABC: a second NSW site, and OpenAI told Sydney four months late
Judge: Expecting Google Traffic Is Not a Deal The Verge: Chegg also lost its claim that indexing forces content into AI answers
Newsom Now Wants a Kill Switch in Law TechCrunch: he still vetoed the smart-glasses recording bill as too broad
NEWTrump's AI Czar: His Spy Chief CNBC: Clayton twice dodged the job question and argued against any pause
The FalloutJobs, money, daily life
NEWNurses Say Palantir's Scheduler Breaks Shifts
Google's Classroom AI, and Falling Scores NY Post: 70% of teens use AI for schoolwork as test scores fall
Micron Sells Memory to 2030, Sight Unseen The Verge: Samsung added $100 to most of the Galaxy S26 line
ChatterOpinion, drama and noise
NEWChatGPT Will Put the Dress on Your Selfie Android Authority: it also finds lookalikes from a photo you upload
Ghost Note Three Years for an AI Crocodile
GHOSTS IN THE MACHINE — AI news without the engagement trap.
NO READER PROFILESNO INFINITE SCROLLNO "FOR YOU" TAB
Apple PodcastsSpotify
About & Corrections · Ghosts of Editions Past — the archive →