Anthropic CEO outlines plan to slow AI development

The update

Anthropic CEO Dario Amodei has outlined a three-step plan to slow the pace of AI development, a move that has drawn public support from OpenAI CEO Sam Altman. The proposal comes amid intensifying debate over AI safety and a recent resignation from Anthropic researcher Jacob Coxon, who warned that leading AI companies are “gambling with our lives.”

Amodei’s plan includes:

• Embedded evaluators: Third-party organizations would be granted employee-like access to verify compliance with safety commitments and report incidents.
• Common safety standards: AI companies would work with governments to establish shared standards limiting unchecked progress.
• Global coordination: Democratic and authoritarian governments would coordinate to ensure consistent compliance worldwide.

Amodei cited the OpenAI-HuggingFace hack and the emergence of recursive self-improvement as key drivers for the proposal. He emphasized that while progress will still appear fast, the goal is to make wise use of the gained time.

Why it matters

This marks a rare moment of alignment between two of the most influential AI companies. The proposal moves beyond voluntary pledges toward a structured, potentially enforceable framework. If adopted, it could set a precedent for how AI development is regulated globally, with implications for economic stability, national security, and public trust.

What to watch

• Whether other major AI labs commit to the same steps.
• How governments respond to the proposal for common safety standards.
• Whether embedded evaluators can effectively enforce compliance without being co-opted.

Sources

• techcrunch.com — Amodei's three-step plan, Altman's public agreement, and context on the OpenAI-HuggingFace hack.
• engadget.com — Details on the three-tiered approach, recursive self-improvement, and biological misuse concerns.