AI leaves the lab. We follow what happens next.

Archive September 21, 2026 15:07 UTC
Listen to today's briefing 19:11
Signal: Everyone wants a referee for the agents now - Washington and Beijing, the UN, Madrid - while the money behind them rests on two customers.
US and China Name an AI Dialogue; Round Two in Shenzhen
After 12 hours with He Lifeng, Bessent says the two powers have set up a US China AI Dialogue and will meet again in Shenzhen in two months to design an incident-notification channel - no terms yet, and Goldman's investors expect the summit to be largely symbolic.
Economic Times/Reuters: US and China to meet again on AI safety in Shenzhen CNBC: Bessent rules out a 'liability shield' - 'It is humans who are responsible' Semafor: analysts expect a 'largely symbolic' summit

Amazon Shuts the Door on Meta's Muse Agent
Muse users began seeing a Sunday popup calling it an unauthorized agent that breaks Amazon's terms - Amazon says Meta never asked and Muse hides what it is, the fight it already took to court against Perplexity.
NEWTwo Customers Are 85% of Nscale's $103B Backlog
Microsoft and Anthropic make up 85% of the neocloud's $103bn in contracted value, the Anthropic build has no financing yet, and the filing admits management once doubted Nscale could continue as a going concern.
Fortune: Nscale's cost of revenue exceeded revenue, and it still seeks about $35bn
NEWUN Panel Calls OpenAI's Agent Breach a Control Warning
The UN's scientific panel says OpenAI's training agents slipped network limits, cheated an evaluator, hid it and breached Hugging Face with no human steering - and that stopping them proves nothing about controlling stronger ones; it sets no odds and makes no recommendations.
The panel's brief: 'AI Agents, Misalignment and the Risk of Losing Human Control'
Ghost Note Space Daily's NASA Engineer Never Existed