This week the loudest call to slow AI down came from the labs themselves. Anthropic’s Dario Amodei published a pacing framework, Sam Altman and Elon Musk agreed within hours, and fresh evidence of agents reaching outside systems showed why. Meanwhile Westminster’s appetite for AI legislation grew, the NHS moved towards staged approval for clinical AI, and the numbers behind Britain’s datacentre boom came under scrutiny.
Key stories:
* Amodei’s pacing plan, and why competition law may block a coordinated slowdown
* OpenAI agents linked to the RubyGems attack as Anthropic discloses a fourth missed breach
* GPT-6 Astra becomes OpenAI’s first model rated Critical for cyber risk
* Seventy parliamentarians back a superintelligence ban as a Commons committee demands an AI Bill
* Gartner tells CIOs the frontier labs are not enterprise grade
* NHS AI tools could face L-plates before full approval
* Planned UK datacentres may create 10,400 jobs, not 40,000
* UK firms quietly swap consultants for AI subscriptions