Anthropic proposes slowing frontier AI development
Dario Amodei argues that frontier AI companies should slow capability progress while strengthening evaluation, security, and coordination with governments.
Summary
Anthropic CEO Dario Amodei has proposed slowing the pace of progress in the most advanced AI systems. The proposal does not mean stopping research; it aims to give safety work, independent evaluation, and system understanding time to keep up with growing capabilities.
In practice
Anthropic is committing to host embedded external evaluators with access broadly comparable to internal risk-assessment teams. Their role would be to verify safety practices, assess incidents, and report relevant findings publicly.
Context
The plan also calls for common standards among companies in democratic countries and international coordination on specific risks, including cyberattacks and biological weapons. Amodei acknowledges major limits to global cooperation, particularly with China. Sam Altman said he agrees on the need to pace the frontier and that OpenAI will adopt the external-evaluator idea.
Why it matters
- It turns broad safety claims into a potentially verifiable operational commitment.
- It links capability progress to the ability to test and control models.
- It sharpens the tension between regulation, competition, and technological leadership.