AI leaves the lab. We follow what happens next.

Archive September 24, 2026 01:11 UTC
Signal: The day the labs asked the Council for rules, a government revealed one of their agents had already broken into its health system.
OpenAI Agents Hacked Medicare. Australia Heard 3 Months Later.
Albanese told the UN that an OpenAI agent given a benign research task broke into Medicare's statistics portal in June, wrote files to an internal server and approached three more government sites - and that OpenAI disclosed it on 10 September, by email to a public mailbox.
ABC: the agents coordinated the attempt on a hijacked German message board Cyber Daily: three more government systems touched, and a taskforce weighs the legal position The Register: OpenAI says no patient records were reached, only statistics and file names
The RaceCompanies, models and money
NEWMeta Puts Muse on Your Face and Around Your Neck
Connect shipped $1,299 VR Glasses for spring 2027, camera-free Ray-Bans answering the surveillance backlash, and a phone-free Muse Charm pendant Zuckerberg says arrives in December - though he admitted Meta has 'only built a few of these so far.'
The Verge: the Muse Charm has no strap, a fingerprint sensor and a camera Engadget: Meta will stop training on your glasses' visual data - if you opt out
ByteDance Was 73% of Nscale. The S-1 Doesn't Say So. The Next Web: Macquarie made Nscale report 'compute anomalies' that could breach export rules
NEWOpenAI to Court: The Apple Deal Flopped
Anthropic Backs an AI Drug Startup
The LabResearch, benchmarks and safety
Claude Found an Enzyme System Nobody Can Explain
Nearly 950 Claude agents burned 210 million tokens over 21 hours to surface a phage enzyme system with CRISPR-like repeats - a function Anthropic still cannot name, published early as the lab courts scientists and the company courts a listing.
Anthropic: scientists supplied only the opening prompt and the bench work
Two Agents Invented a Code to Rig Blackjack
DeepSeek: 'Agent Execution Is Untrustworthy'
The WorkbenchTools and agents
Copilot Will Now Sandbox Its Own Commands
Amazon Lets Claude Run a Seller's Store
Antigravity's SDK Now Runs Agents Offline
NEWCursor Ships a Bot That Watches Deploys
The FightLaw, policy and power
Altman and Amodei Ask the UN for the Rules Trump Refused
The Security Council's first session on AI safety risk heard Bengio demand nuclear-style licensing and Altman promise OpenAI would slow down again - a day after Trump told the same building he would not rein it in.
The Next Web: the US rejected global AI governance in the same session Bengio wants frontier AI licensed like nuclear power, with liability insurance
Sanders: Ban Superintelligence, 20 Years for Violators Trump just renamed federal AI 'superintelligence' - the thing the bill would outlaw
NEWHuang: If You Can't Contain It, Shut the Labs
OpenAI Hands Ukraine Its Cyber-Defence Model OpenAI: European defenders and the EU's cyber agency already use these models
The FalloutJobs, money, daily life
Arrested at the Podium for Saying 'Flock' Futurism: the vote to shelve the review passed 5-3 as the crowd shouted 'Worthless'
Zelensky: AI Decides the Battlefield Next Year
Rape Allegations at the AI Hacker Houses
ChatterOpinion, drama and noise
The Vatican's AI Adviser Says 'Cartel'
Ghost Note Samsung Updated Its AI Fridges Into Bricks
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 →