AI leaves the lab. We follow what happens next.

Archive September 22, 2026 15:11 UTC
Listen to today's briefing 12:16
Signal: Governments move on the machines: Centcom retools its AI targeting, Beijing probes its labs' Claude traffic, and banks want rules before agents shop.
After 120 Children Died, Centcom Rebuilt Its AI Targeting
Officials say Centcom changed how it vets targets, added open-source feeds to track civilians and made dozens of upgrades to Palantir's Maven after the Minab school strike - the Pentagon's own review is still unreleased, and some officials warn another Minab remains too likely.

Alibaba's New Chip, a 10-Trillion Model and a 20GW Bet
At Apsara on Tuesday Alibaba said its Zhenwu V900 triples May's chip and set a 20GW cloud target for 2032 as Eddie Wu pitched superintelligence - self-reported specs with no outside benchmark, a day before Xi lands in Washington.
The Register: Wu says AI will do 99.9% of cognition; the US has 37GW under construction SCMP: Alibaba teases a model of up to 10 trillion parameters
NEWBeijing Probes DeepSeek, Moonshot for Feeding Claude
China's internet regulator is examining whether police, military and corporate data reached Anthropic after the US lab said its labs pushed millions of exchanges through Claude to distill it - per The Information, with no sanctions announced.
The Decoder: Xiaomi's MiMo-V2.6 tops open models; Xiaomi is also on Anthropic's list
B.C. Sues OpenAI Over Tumbler Ridge Chats It Won't Hand Over
British Columbia filed in California on Monday, saying the shooter's ChatGPT conversations alarmed OpenAI enough to trigger an internal review and a ban but never a call to the RCMP - and that OpenAI has refused to disclose them; the claims are untested.
CJDC-TV: OpenAI banned the shooter's first account; a second one got around it
Ghost Note OpenAI's Trainers Fired for Using AI