Anthropic, the AI startup valued at $900 billion, urged a worldwide “brake pedal” on frontier models in a blog post Thursday. Founder Jack Clark told the BBC the industry lacks a safety mechanism to curb recursive self‑improvement, where systems could upgrade themselves without human oversight. The call arrives as the company prepares for a high‑profile IPO and attract major institutional interest.

Anthropic proposes an AI version of a nuclear non‑proliferation treaty, requiring several well‑funded labs across nations to agree on a coordinated pause and to verify compliance. Its internal Anthropic Institute would develop verification tools. Critics, including former AI czar David Sacks, dismiss the plan as fear‑mongering, while investors worry a slowdown could crush stock valuations built on rapid growth and could trigger regulatory hearings.

The proposal could reshape capital flows to AI firms, whose IPOs target trillion‑dollar market caps. If regulators adopt a global brake pedal, funding rounds may dry up, pressuring valuations like OpenAI’s and Anthropic’s. With the U.S. and China eyeing the same technology, the debate pits national security against shareholder returns, and the market will feel the pressure immediately.