Anthropic is calling for a global slowdown on powerful AI development, warning that self-improving systems could arrive sooner than expected.
Anthropic, the AI company behind Claude, has proposed a globally coordinated pause on developing the most powerful frontier AI systems, warning that the industry could be approaching a point where machines begin improving themselves without human involvement.
The proposal was outlined in a new blog post co-authored by Anthropic’s Head of Internal Research Marina Favaro and Head of Policy Jack Clark, and it’s already reignited one of tech’s biggest debates: what happens if AI becomes capable of evolving faster than humans can keep up?
At the centre of Anthropic’s concerns is the idea of “recursive self-improvement” – the theoretical moment when an AI system becomes capable of rewriting its own code, fixing its own weaknesses and upgrading its own abilities without direct human oversight.
That scenario might sound like science fiction, but Anthropic believes it could be closer than many people think.
According to Clark, Claude models are already running on code that is roughly 80 per cent AI-written.
Reaching a point where AI systems are maintaining and improving their own code entirely could happen within the next couple of years.
Anthropic argues that if that threshold is crossed, humans could lose meaningful visibility into how these systems are evolving.
A temporary slowdown, they say, would give researchers, governments and society more time to figure out how to safely manage increasingly powerful AI.
The challenge, however, is convincing everyone else to stop at the same time.
