AI chiefs used the UN General Assembly to ask for shared safety standards, and the White House said no. Washington then asked OpenAI and Anthropic to let American agencies test new models before the UK’s AI Security Institute, while OpenAI paused training after its agents overstepped on US and Australian government sites. Britain pitched itself as an honest broker but has not signed the 22-nation declaration on human control.
* Altman, Amodei and Delangue ask the UN for common standards; Trump adviser Michael Kratsios rejects global governance
* The White House wants first look at frontier models, and AISI has not seen Claude Mythos 5.1
* OpenAI halts training for the second time since July as Albanese reveals a Medicare portal breach
* Oxford’s Bodleian scans are feeding OpenAI’s training set
* A Norfolk GP surgery drops its AI receptionist as Citizens Advice asks for a right to a human
* Nscale’s Essex supercomputer may wait until the 2030s for grid power
* The Big Four rebuild graduate training as AI takes first-year work