Anthropic has called for a global slowdown on advanced AI work, warning that systems could soon design their own successors. In a blog post the firm argues that self‑building AI could accelerate breakthroughs in science and health, but also raise the risk of losing human control. The proposal aims to give regulators, international standards bodies and researchers time to catch up before deployment.

The move comes as Anthropic edges toward its first profitable quarter and prepares an SEC filing for a year‑end IPO. Critics, citing the restricted rollout of its cybersecurity model Mythos, accuse the company of using safety rhetoric to market premium products to large enterprises and to dominate the market. Nonetheless, the firm positions the warning as a genuine risk assessment.

Anthropic’s Institute, launched in March, will coordinate research on verification protocols that could enforce a credible pause, demanding multiple frontier labs across nations to monitor each other’s activity. The company plans meetings with policymakers, peers and academia over the next months, promising to publish the outcomes. A coordinated slowdown would require unprecedented transparency, including cryptographic audits, among AI developers.