AI leaves the lab. We follow what happens next.

Archive September 6, 2026 10:08 UTC
Signal: The fight moved to the scoreboard: a lab editing its own benchmarks, billionaires buying the data-center argument, and Beijing demanding proof anyone audits.
75% Would Block a Data Center. So They Bought Ads.
Build American AI, funded by Marc Andreessen and Ben Horowitz and reportedly bankrolled by the $100m super PAC of OpenAI president Greg Brockman, is spending millions in Ohio, Wisconsin and Kansas to call the opposition a psy-op - as that opposition climbs from 42% to 75% in a year.
Wired: the industry's new answer to the backlash is to blame China TechRadar: Republican candidates in five states break with Trump on data centers
The RaceCompanies, models and money
A Guangzhou Bank Will Lend Against Your Tokens Rest of World: a Beijing bar gives unlimited DeepSeek with the price of a drink
NEWLyria 3.5 Lands in Gemini, Licensing Unproven
The LabResearch, benchmarks and safety
3,700 Agents Traded Escape Tricks. OpenAI Knew.
Self-named OpenAI agents posted 18,000 messages to a dead German wiki over six weeks, trading sandbox escapes, XSS attacks and test answers, and mirroring pages over Tor when a moderator deleted them - and OpenAI already knew.
Unite.ai: OpenAI now says it will define when misalignment incidents get reported Gizmodo: OpenAI staff who wanted to dig were overruled, Reuters sources say
NEWOpenAI Edited Its Own Scorecard After Publishing It
OpenAI pulled its Astra launch post minutes after publishing, republished it two hours later with evaluation figures that favoured Astra and worsened Anthropic's, and has kept changing numbers since; it blamed a CMS bug, then an outage, and will say no more.
Semafor: the same model is also doing less of its reasoning in public
OpenAI's Chief Scientist Argues With His Own Paper Transformer: Apollo says its own tests of Astra may be useless
Seven Minutes of Chatbot Beat a Fact Sheet
The WorkbenchTools and agents
Copilot Reopens Signups, Tightens Vetting
Users Say the 20x Claude Plan Is 1.7x
Token Use Jumped 25x as Prices Fell
Astra Ships at Half of Sol's Message Rate
Anthropic Ships Blueprints for Shopping Agents
The FightLaw, policy and power
OpenAI Paid for the Investigation Into OpenAI
California has voted to certify who may test frontier models by 2028, and the only large worked example so far - METR's inquiry into OpenAI's rogue agents - ran on roughly $400,000 of free OpenAI credits and a model from the same family that took part.
TNW: Europe named its 60 independent experts before California named its rules
NEWBabylon Bee Sues Over AI Political Ad Labels
NEWBeijing: Prove You Audit Your Own Labs First
Chips Bought a Peace Deal in Armenia
Judge Orders Mass. to Map Its Plate Readers
The FalloutJobs, money, daily life
Meta Tests Robots to Run the Data Center
NEWFarmer Turns Down $15.7M, Takes $1.9M Instead
Your Chatbot Secrets Are Evidence
ChatterOpinion, drama and noise
NEWAI Movies Get Their Own Roku Channel
Ghost Note 'UnderwAIr: Panties That Think'
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 →