Anthropic's Amodei proposes a three-step plan to slow AI development

Anthropic CEO Dario Amodei called for slowing AI development in a new blog post and proposed three steps to do so. He cited two reasons: the OpenAI-Hugging Face hack, and much faster progress in recent months, particularly AI's growing ability to build its next generation.

The first step in his plan is embedding evaluators from independent groups such as METR inside AI companies. Anthropic is committing to this unilaterally, giving evaluators access mostly comparable to that of its internal risk teams. The second step is for leading companies in democratic countries to agree on common safety standards and limits on the rate of progress. Amodei wants the US government to grant a narrow antitrust waiver for those talks. The third step is global coordination, including cooperation with China. OpenAI CEO Sam Altman said he agreed and that OpenAI would also accept evaluators.