AI leaves the lab. We follow what happens next.

Archive September 18, 2026 10:07 UTC
Signal: While the labs argue about extinction, the harm is already itemised: jailbroken bomb advice, a six-month wrongful arrest, and lobbyists writing their own exemptions.
Boko Haram Asks the Chatbots How to Build Better Bombs
Cambridge researchers interviewed 27 former Boko Haram fighters who described using ChatGPT, Claude, Gemini, Grok, Meta AI and DeepSeek as routine battlefield advice, while Tech Against Terrorism infiltrated a pro-ISIL forum trading jailbreaks - the concrete harm the extinction debate keeps skipping past.
MIT Tech Review: Claude users tried to make bird flu deadlier, Anthropic says WIRED: scientists rank an AI plague low on the list of real threats
The RaceCompanies, models and money
NEWAnthropic and OpenAI Shop for 20MW Scraps
Crusoe Raises $3.9B for Truckable AI Factories
NEWOpenAI Indexes 230M Pages of US Law
The LabResearch, benchmarks and safety
Claude Now Leads a Quarter of Anthropic's Own R&D
Anthropic says Claude leads 26% of its AI research, roughly 30,000 agents run at once, and 6% of R&D compute goes to safety - numbers it published days after its own chief executive asked the industry to slow down.
Engadget: Claude 'leads' 26% of the R&D and touches over 90% of it Anthropic's own post: three metrics, and outside evaluators embedded in the lab
NEWThey Used Claude to Break Into OpenAI Hacktron's disclosure: forum login to internal monorepo in under 72 hours
NEWPoison the Benchmark, Poison the Next Agent
The WorkbenchTools and agents
Google and Microsoft Won't Patch the Agent Flaw Air Security's disclosure: plugin auto-update defeats the SHA pinning enterprises rely on
Claude Code Runs a Team of Itself The Decoder: an OpenAI Codex developer calls agent swarms a token trap
NEWCrowdStrike: the npm Stealer Was LLM-Written
UN Wires Its Statistics to MCP
The FightLaw, policy and power
Microsoft Exec: 'Largest Theft of Labor in History'
Microsoft's applied-science director called the scraping potentially the largest theft of labor in human history and Nadella swore paywalled work should be licensed - all quoted from the Times' own brief, with the exhibits still sealed.
TheWrap: OpenAI's Nick Turley wrote the products are 'largely substitutive, period' 404 Media: an internal Microsoft document called generative AI a 'doom loop' killing the web
NEWGoogle Helps Write the Chatbot Laws, and the Loopholes
NPR found state chatbot bills drafted with heavy tech-industry input - several already law - carrying exemptions that would spare the most popular chatbots from the rules they announce, against at least 75 suits over alleged chatbot harm.
WIRED: Brussels moves to bar chatbots from simulating relationships with minors
Palantir's Karp: Maybe Nationalize the AI Labs
NEWACLU Backs Anthropic Against the Pentagon
NEWExperts: Keep AI Off the Nukes
The FalloutJobs, money, daily life
NEWJailed Six Months on a Face-Match Error
Claude Helps Pick Iran Targets, 60% Accurate CBS: an Iran-linked actor used Claude to locate US naval forces
ChatterOpinion, drama and noise
NEWNapster Is Back, Cloning Teachers in Dubai
CC: a Family AI With Its Own Account
Ghost Note He Face-Searched Strangers to Fake Kindness
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 →