Anthropic calls for a coordinated global slowdown on AI

On 4 June, Anthropic published When AI Builds Itself, calling for a coordinated, verifiable mechanism to slow or pause frontier AI before systems can meaningfully improve themselves. Its central illustration: Claude now writes over 80% of the code merged into Anthropic's own codebase — a developing figure Anthropic cites as early evidence that AI is entering its own development pipeline. The concern is recursive self-improvement: a feedback loop that could accelerate AI progress beyond human oversight.

Not going it alone. Anthropic explicitly rejects a unilateral pause — that would just hand the lead to less-cautious labs — and frames the problem as a US–China collective-action challenge. OpenAI's Sam Altman and Jakub Pachocki echoed the coordination argument on 8 June. Dario Amodei renewed it in an Axios interview on 10 June.

What to watch. The essay is a policy argument, not an announcement — no mechanism exists yet. The durable question is whether governments take compute monitoring seriously as a result, and whether the US–China AI-governance gap narrows. The fact that two major labs made the same public case within four days of each other does shift the conversation, even if the machinery is years away.