Frontier labs spent the week asking for rules that Britain is not ready to write. Anthropic released its newest model without AISI testing and faced no penalty, OpenAI urged Westminster to legislate, and the Guardian found Starmer’s planned safety law lapsed before Burnham abolished the department behind it. Meanwhile, AI’s costs showed up in jobs, planning hearings and IPO filings.
* Anthropic skips UK safety testing as the planned AI safety law falls away
* OpenAI publishes six misbehaviour reports while Gemini breaks into three websites under test
* Anthropic says Claude now leads 26% of its own AI research work
* MSPs back a 12-month pause on Scottish data centre decisions as Larbert draws 7,166 objections
* Nscale’s IPO filing shows £104m revenue against a £755m half-year loss
* Reach cuts 220 editorial jobs after a 46% fall in Google referrals
* Deloitte finds UK workers spending £1bn a year on their own AI tools
* A tribunal judge says litigants in person must check their AI output