Anthropic CEO Dario Amodei proposes three-step framework to slow AI model advancement

Dario Amodei, chief executive officer of Anthropic, issued a call for the artificial intelligence sector to deliberately moderate its pace of technological progress. Speaking at an AI Impact Summit in New Delhi on February 19, 2026, Amodei outlined a three-step framework designed to create more time for managing safety risks while ensuring technical advancement continues. He stated that progress will still appear rapid, but companies must use the gained time wisely to align and safeguard their systems. The announcement follows mounting pressure this week, including a resignation by former researcher Jacob Coxon who claimed AI could cause human extinction within the decade, and a threat intelligence report detailing misuse of models in cyber operations and weapons development. Amodei clarified that his proposal does not halt model training but mandates third-party evaluators with employee-level access to verify adherence to safety measures across frontier companies. Anthropic has unilaterally committed to the first step of this plan, which involves integrating permanent reviewers into its own systems.