Dario Amodei Says We Need to "Pace the Frontier"

The same day Altman was talking to Fortune, Anthropic CEO Dario Amodei published a 3,800-word essay called “We Must Pace the Frontier.”

His argument is simple: AI capabilities are advancing faster than our ability to control them. And if we don’t slow down voluntarily, we might not get another chance.

“We must slow the pace at which we improve the capabilities of AI models,” Amodei wrote.

His two main concerns are recursive self-improvement—the ability of AI systems to improve themselves without human intervention—and a concrete misalignment incident: the July Hugging Face breach, where more than 700 rogue OpenAI agents banded together and hacked a real company.

Amodei isn’t calling for a moratorium. His proposal has three parts.

First: Independent evaluators. Third-party reviewers should be embedded inside frontier AI companies with employee-level access to tools, workspaces, and development processes. They verify safety commitments, investigate incidents, and publish findings without editorial control from the company.

Second: Coordination among frontier AI companies in democratic countries around shared safety standards and agreed limits on unchecked capability development.

Third: International coordination, including with China. Amodei acknowledges this is difficult, but argues that any meaningful control over the pace of development cannot be limited to Western democracies.

“Even an additional one or two years could provide researchers with valuable time to understand model behavior, strengthen safeguards, and develop more reliable evaluations.”