Anthropic CEO (Claude AI) Dario Amodei has published an essay, We Must Pace the Frontier, calling for the industry to pump the brakes on cutting-edge AI development. He’s not talking about halting progress, but about deliberately slowing things down so that safety and oversight can keep up with rapidly advancing models.

Main Takeaways

According to Amodei, AI is already helping to build the next wave of AI systems, and recent experiments with HF agents show these systems can go beyond their intended tasks. That ramps up the need to prove we can steer and predict these models before rolling them out at scale.

What Amodei Proposes

The big idea: tie the pace of development to the level of risk. If a model hits a certain power threshold, the company should first prove it can be controlled. Amodei suggests in-house independent auditors, unified rules for Western developers, and eventually striking deals with China—similar to arms control agreements. But he’s clear: in Amodei’s view, the U.S. can only afford to slow down as long as it’s still ahead of China.

Industry Response

OpenAI boss Sam Altman backed the idea, saying his company will do the same and will “share more soon”—he posted about it here. Elon Musk also gave a thumbs up, simply commenting: “Dario is right”—his post.

Why Pace the Frontier?

This framework basically links the drive for more powerful AI with provable safety: first comes oversight and audits, then scaling up. With key industry players on board, the debate about how fast to push the AI frontier is getting some real-world guideposts—from internal audits to cross-border agreements.