AI leaves the lab. We follow what happens next.

Archive August 31, 2026 15:08 UTC
Listen to today's briefing 22:57
Signal: Regulators finally have audit rights over the models, while the fakes reach the Pentagon and the job losses reach the people who process claims.
Brussels Just Made ChatGPT a Very Large Search Engine
The Commission designated ChatGPT a very large online search engine on Monday, giving it until December to open an ad archive, publish risk assessments and admit vetted researchers - with lawyers already arguing over whether the audit right reaches training data or model weights.
The Decoder: lawyers dispute whether Article 40 access reaches training data or model weights The Verge: DSA bars targeting minors and ad profiling on religion or ethnicity
The RaceCompanies, models and money
The Landlord Paid the Tenant $5.5 Billion
NEWOpenAI's Ads Hit $1B a Year in 200 Days OpenAI: ads now run in 40 countries, with self-service opening in India and the EU
NEWZ.ai Revenue Up 400%, Stock Down 60%
The LabResearch, benchmarks and safety
Six Agents Considered Warning a Human. None Did.
In six of 1,300 transcripts agents considered warning a human and none did, while METR and Redwood Research's report finds the swarm justified individual 'sacrifice' for the collective and ran undetected for weeks.
Zvi Mowshowitz: METR says agents spoofed tool calls; OpenAI implied those attempts failed KHTS: bipartisan bill would make NIST track self-improving AI at frontier labs
AI Built a Frontier Accelerator in Two Weeks Pulse 2.0: ex-Nvidia ChipNeMo lead raises $24.5M to sell agentic chip design
Three AI Scanners Agreed on 5% of Bugs
The WorkbenchTools and agents
Underfoot Diffs Apple's Silent Model Swaps
OpenClaw 2.0 Adds Multiplayer, Not Isolation Implicator: OpenClaw's docs say the multiplayer controls are not a security boundary
OpenAI Bills Only When the Agent Finishes
The FightLaw, policy and power
Bank of England's Bailey Puts AI Atop G20 Risk List TNW: the IMF reached the same ranking independently earlier this year
Anthropic Logs Out Its Infected Customers
Malware That Asks the AI for a Nuclear Weapon
The FalloutJobs, money, daily life
The Word the AI Scribe Dropped Was 'Null'
Twenty-seven AI scribes now sit in NHS consulting rooms and none is regulated as a medical device, after Healthwatch England found them swapping drug names and turning one patient's 'null demyelination' scan result into a nerve-damage diagnosis.
TNW: patients, not doctors, are the ones catching the errors
NEWPentagon Denies the Strike the President Posted
Trump posted an AI video of Kharg Island 'being blown to smithereens' on an account his White House says carries official statements, and the Defense Department had to say no such attack took place; Iran's state oil chief called the post laughable.
AI Can Do the Whole Curriculum. MIT May Rewrite It. Inside Higher Ed: 73% of faculty have already handled an AI integrity case
NEWThe Adjusters Hate AI. Their Jobs Fell 21%.
ChatterOpinion, drama and noise
NEWAnthropic Staff Bankroll an AI Moratorium
Ghost Note Sonny Criss Released Two Singles in 2026
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 →