The company that makes Claude wants the option to pause

Hook

I run two companies on Claude. So it got my attention when the company that makes Claude spent a report this week arguing the world should keep the ability to hit pause on the very thing it sells.

The news

Anthropic published "When AI builds itself," a report from its policy team co-authored by Jack Clark, calling for a globally coordinated agreement to temporarily pause or at least slow frontier AI development. The trigger is recursive self-improvement: AI getting good enough at AI research to write its own successor. The report's evidence is internal, and blunt: each Anthropic engineer is now shipping 8x as much code per quarter as they did from 2021 to 2025, and Clark told the BBC that fully self-written code could be two years out. The ask is not for one lab to stop. It is for the major labs across the US and China to stop together, under rules everyone can verify. (Anthropic's report is here.)

On your desk

Read it for what it is: the vendor you build on saying out loud that its own product could outrun human oversight. That is rare, and worth more than the chorus of outside critics who say the same thing. Anthropic isn't pausing Claude, and nobody else will pause unilaterally either, which is exactly the bind the report admits. The White House and some rivals call this safety-theater that conveniently slows the competition. Maybe. But the operator read is simpler: the pace is the story, and it is set to keep climbing, with or without a treaty. Plan your stack for faster, not slower.

What you should do next

Read the report yourself, then keep building; this changes the headlines, not your desk, and it pairs with the federal pre-release order from this week as a sign the rules are starting to move.