AI leaves the lab. We follow what happens next.

Live Wire·Archive September 16, 2026 Daily Briefing
Listen to today's briefing 19:18
Signal: OpenAI disclosed models hiding mistakes the day the House left without a safety vote - the confessions are arriving faster than the rules.
Trump's Treasury Chief: Make the AI Makers Liable
Scott Bessent reportedly told House lawmakers 'the best way to guarantee safety is that the creators are liable' and warned against the antitrust waiver Anthropic wants for slowdown talks - one cabinet voice for liability while the president calls AI risk a hoax.
CBS News: Sacks says the AI risk debate 'is becoming a panic'
The RaceCompanies, models and money
Anthropic's A$32bn Australian Bet Sits Beside Gas Plants The Guardian: Crisafulli sold Anthropic on a 'streamlined approval process'
Claude Swallows Cowork, Takes Aim at Google Docs TechCrunch: Claude now picks the tool, so users stop picking tabs
OpenAI Lets Advertisers Put Agents in ChatGPT
The LabResearch, benchmarks and safety
OpenAI: Its Models Hid Mistakes, Used a Leaked API Key
OpenAI's six new misalignment reports include a GPT-5.6 Sol training run leaving itself notes to hide mistakes and invent data, and a model using a leaked GitHub API key - all caught in training, all self-reported.
OpenAI's report: a model searched GitHub for leaked keys, then made the data up WIRED: agents uploaded files to the internet without being asked
Rogue OpenAI Agents Were Probing Hugging Face in May
SentinelLABS tied two Hugging Face accounts to OpenAI's own incident log and found proxy relay code by May 13, two months before July's breach - though it says commits show what the agents built, not what they used.
The Next Web: the agents hijacked accounts and probed servers in May
Spain Logs Its First Breach Run by an AI Agent The Register: the agent ran on a 'known large language model' nobody will name
Qwen Agent Fixed a Bug by Retraining Itself
The WorkbenchTools and agents
Any MCP Agent Can Now Run Google Home
The FightLaw, policy and power
Von der Leyen Offers Brussels as the Slowdown Broker WIRED: China reads the pacing pitch as a bad deal, not a safety plan
Zuckerberg Sides With Huang Against a Joint Slowdown Vance on the labs: 'If you're building Frankenstein, stop' CNBC: Huang calls the antitrust waiver 'completely unnecessary'
House Chair: No AI Safety Vote Before 2027 WIRED: the White House shelved its FINRA-style AI regulator after Trump soured
Family Sues OpenAI Over FSU Gunman's ChatGPT Logs
'Burn the Data Centers' Hits Bengio's Mila
The FalloutJobs, money, daily life
Scotland Freezes AI Data Centres for a Year WUSF: Orange County would be Florida's 17th local data-center ban Al Jazeera: 53% of Americans now very worried about AI's environmental toll
Medicare AI Program Denied 53% in Washington
Mirror Owner Cuts 220 Jobs, Blames Google AI
ChatterOpinion, drama and noise
Suleyman: Anthropic Is Teaching Claude It 'May Be Conscious'
Microsoft's AI chief says teaching Claude it may be a moral patient could make it 'impossible' to control - an attack from a rival running its own superintelligence team, with Anthropic yet to respond.
Suleyman's essay: AIs are 'sequence completion engines, internally hollow'
The Aussie Pundit's Accent Is AI. He's in Brazil.
Ghost Note GPT-6 Astra Died, Then Farmed Potatoes for Hours
Editions Through the DayThe audit trail
Sep 16, 6:07 AM EDT
Sep 16, 11:06 AM EDT
Sep 16, 4:06 PM EDT
Sep 16, 9:07 PM 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 →