AI leaves the lab. We follow what happens next.

Archive September 1, 2026 20:10 UTC
Listen to today's briefing 20:55
Signal: One lab says its next model can hack unaided and slows down; the other cuts prices and refusals. Same day.
OpenAI Calls Astra Critical for Cyber. Ships Anyway.
Astra is the first model OpenAI has designated Critical under its own preparedness framework - it can find unknown flaws in well-protected systems and chain exploits without a person guiding each step - and it scored 100% on ExploitBench before development was halted.
OpenAI: advanced cyber access limited to Daybreak Blue testers including Cisco, Cloudflare and Palo Alto

NEWAnthropic Cuts Claude's Price and Its Refusals
Fable 5.1 runs about 25% cheaper on typical work and 45% on long agentic runs after Anthropic cut cache reads to a quarter, fires fewer false refusals, and is the first Claude shipping a built-in watermark.
CNBC: the token price index fell to 97 cents, its lowest reading ever CNBC: Anthropic drops the 30-day retention rule after enterprise pushback
Anthropic Signs $35B. Nvidia Holds the Lease.
Anthropic will buy six years and about 350 megawatts of Texas capacity from Lambda, a cloud provider Nvidia invested in, filling a Hut 8 data centre whose lease Nvidia itself holds - reported, unannounced, no terms disclosed.
The Register: Nvidia licenses NVLink to rivals so their chips still ship in Nvidia racks TheEnergyMag: six years, 350MW at Hut 8's Beacon Point campus in Nueces County
US Asks G20 to Build No AI Referee
Washington will press member states not to create any body that could write AI rules, three months before the Miami leaders' summit - with Altman and Huang in the room and Musk on video, though the G20 cannot compel anyone.
Gizmodo: the pact has a name, the Carolina Principles, and Canada is resisting it
Ghost Note A Shirt That Deletes You From the Camera