Safety

Anthropic Warns AI Is Beginning to Build Itself and Urges a Verifiable Global Pause

The company says a coordinated slowdown would give regulators and researchers time to study risks from recursive self‑improvement under rules that can be independently checked.