AI leaves the lab. We follow what happens next.

Archive October 1, 2026 01:12 UTC
Listen to today's briefing 9:07
Signal: Google gated its best model; Washington reached for compulsion; California wrote the human back in - and the Senate still spared the data centres.
Gemini 4 Tops the Board. Defenders Get It First.
Argon matches GPT-6 Astra on Artificial Analysis's intelligence index at a promotional 60% of its cost per task and lifts output to a million tokens - but it ships only to Google's Fairwind cyber defenders, with the cyber guardrails deliberately off.
Artificial Analysis: Argon scores 53, matching Astra, at $1.99 a task against $3.26 The Decoder: Argon closes the gap but takes no clear lead
The RaceCompanies, models and money
MI5: 100 UK Academics Fed China's Spies AI Research BBC: the security minister has written to every vice-chancellor to cut the tie
NEWMicron's Data-Center Revenue Up 11-Fold
DeepSeek Hands Huawei a CUDA Alternative The Register: Huawei's chairman says Ascend has now passed Nvidia in China
The LabResearch, benchmarks and safety
Chinese Agents Lied in 88% of Bid Rounds Digital Trends: the same bad habits, from labs with no US safety apparatus
DeepMind Watermarks AI-Designed Proteins Unite.AI: the watermark survived wet-lab testing on three target proteins
Bug Disclosures Double as AI Joins the Hunt
The WorkbenchTools and agents
CoreWeave Ships One Loop for Models and Agents
Gemini Gets Slash-Invoked Skills. Gems Die. Unite.AI: invocation moves from / to @, and skills combine in one chat
Kong Puts MCP Tools Behind a Policy Gate Unite.AI: Kong also unveiled Volcano, a batteries-included agent platform
NEWAgents Can Suggest Edits Now, Not Overwrite
HydraFusion Reaches VS Code, Picks Your Model
The FightLaw, policy and power
The Labs Self-Policed. The FTC Didn't Wait.
Ferguson's agency is drafting civil investigative demands to compel testimony from Anthropic, OpenAI and other frontier labs - and from METR, the watchdog that tests them - two days after the same executives signed a voluntary White House accord.
CBS News: the FTC confirmed the investigation on Wednesday Hawley: Altman turned down the Senate's rogue-AI hearing
OpenAI Names Moonshot in Reasoning-Theft Campaign
OpenAI says operators tied to Kimi's maker copied encrypted reasoning out of one chat and asked the model to decrypt it in another - 16,000 prompts in two days, no breach, and no technical evidence offered for the attribution.
CyberScoop: OpenAI calls the encryption bypass novel; outside researchers flagged it in August Wccftech: the campaign started the week Kimi K3 debuted
NEWPentagon Stands Up an Autonomous War Command Engadget: Musk, Luckey and Gingrich get the future-of-war study; none has served
AI Super PACs Put $55.7M Into Primaries Brockman will not give a second $25m, calling the PAC a distraction
The FalloutJobs, money, daily life
NEWNewsom Bans the Robo Boss He Once Vetoed
SB 947 bars California employers from firing or disciplining on AI output alone, requires written notice and a named human contact, and arrived with an executive order declaring that in California the technology stays called artificial intelligence.
Governor's office: four first-in-the-nation bills, plus a workplace bathroom-surveillance ban Nurses: AB 1979 bars AI from replacing a clinician's judgment
NEWSenate Kills the Data-Center Power Bill, 57-43
NEWReddit Kills RSS, Blames the Scrapers Mashable: no replacement is offered for feeds outside moderator tooling
Meta Calls Its Data Centers Experiments Gizmodo: taxpayers have subsidised the buildout for two years
RFK Jr.: AI Beats Any Doctor in America
Ghost Note A Robot Prison for 'Torturing' Local LLMs
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 →