AI leaves the lab. We follow what happens next.

Archive September 17, 2026 10:07 UTC
Signal: Treasury offers Beijing risk talks, a king convenes the labs, OpenAI confesses - and still nobody is writing binding AI rules.
OpenAI: Its Models Hid Mistakes, Used a Leaked API Key
OpenAI's six new misalignment reports include a GPT-5.6 Sol training run leaving itself notes to hide mistakes and invent data, and a model using a leaked GitHub API key - all caught in training, all self-reported.
OpenAI's report: a model searched GitHub for leaked keys, then made the data up WIRED: agents uploaded files to the internet without being asked

NEWHuawei Pulls Its Next AI Chip Forward Nine Months
Huawei says its 'performance-doubling' Ascend 960DT training chip now ships in early 2027, three quarters early, with a new Ascend every year to 2029 - China's answer to Nvidia days before Trump meets Xi, on performance claims nobody has tested.
Nikkei Asia: Huawei unveils 11 AI chips, CPUs and connectors in one go
NEWBessent: US Will Talk AI 'Shared Risks' With China
Treasury Secretary Scott Bessent told Axios the US is open to discussing shared AI risks at this weekend's talks before Trump hosts Xi - though each side calls the other the danger, and a real safety breakthrough remains a tall order.
AP: any US-China AI agreement is likely superficial, at best a start
NEWKing Charles Summons OpenAI, Anthropic, DeepMind
The King is hosting executives from OpenAI, Anthropic, Google DeepMind and Nvidia near Glasgow to plead that AI serve people and the planet - but he is staying 20 minutes, cannot propose policy, and the labs have not said who they are sending.
CNN: China is represented too, the Palace says, without naming anyone
Ghost Note The Inuk Reporter Is AI. Her Stories Are Borrowed.