Overview
- On Saturday, Anthropic chief executive Dario Amodei published a long essay urging AI labs to slow capability advances so safety research and monitoring can catch up.
- Anthropic said it will unilaterally host permanent outside evaluators with employee‑like access to systems, tools, desks, badges and laptops to verify safety practices and report incidents.
- The call followed Anthropic’s recent threat intelligence report on misuse of its Claude models and high‑profile resignations by safety researchers who warned companies were racing toward self‑improving systems.
- Amodei proposed that frontier firms in democracies coordinate common safety standards and ask governments to enable those talks by addressing antitrust limits and export controls.
- Experts and industry figures reacted with mixed support and skepticism, noting commercial pressure, U.S.–China competition and legal barriers could block broad cooperation while the public seeks clearer reporting and enforceable oversight.