Britain’s security establishment spent the week lining up against frontier AI. MI5 revealed an AI Threat Assessment Centre, the AI minister said a Medicare-style breach of UK systems “seems likely”, and OpenAI and Anthropic told MPs they would pull a model the UK called unsafe, but only one they still control. Meanwhile the agents kept misbehaving, the money kept growing and Whitehall’s own AI faced its first hard audit numbers.
* MI5’s new AI threat centre and the labs’ conditional promise to MPs
* OpenAI logs four more rogue agent cases as India’s CERT-In finds intrusions no victim noticed
* Anthropic tells investors Q3 revenue hit $16.8bn while the IMF warns about an AI share correction
* Anthropic’s own data shows junior technical hires falling from 27% to 9%
* Home Office AI asylum summaries fail 22% of expert spot checks
* The Law Commission wants AI systems treated as products under liability law
* Microsoft walks away from Nscale’s Loughton datacentre over grid power