AnalysisPolicySeptember 12, 2026

Anthropic CEO Dario Amodei proposes plan to 'pace the frontier' of AI

Amodei's essay proposes a three-step plan to slow AI capability development, starting with embedded third-party evaluators like METR getting employee-level access to Anthropic's models — a step Anthropic says it is committing to unilaterally. He cites recursive self-improvement and the OpenAI-HuggingFace hack as reasons for caution.

How this story unfolded

1 day · 9 reports · 33 community posts · 42 of 48 shown

  1. Sep 12
  2. Sep 13

More stories today

Open the live feed