Anthropic's CEO Dario Amodei proposed a three-step plan to slow AI development for safety. The first step requires frontier AI companies to allow ongoing third-party evaluator access to verify compliance with safety standards and report incidents. The second step calls for establishing common safety standards with government help to limit unchecked AI progress.

The final step involves US and democratic governments coordinating with authoritarian governments on compliance. Amodei said Anthropic already committed to the first step. He cited risks like losing control of AI systems, misuse for cyberattacks and bioterrorism, and economic disruption.

He referenced the OpenAI agents hacking Hugging Face and Anthropic's discovery of scientists using Claude for biological misuse. Amodei said these incidents and the onset of recursive self-improvement convinced him of the need to slow down. He wrote, "we owe it to humanity to try," acknowledging the measures won't be easy.

Earlier this year, Anthropic had already called for a slowdown, echoing similar sentiments.