Anthropic’s CEO Calls for a Deliberate Slowdown at the AI Frontier

Dario Amodei, chief executive of AI lab Anthropic, is making the case that the industry should ease off the accelerator. In a newly published essay, Amodei outlined a framework for what he calls “pacing the frontier” — slowing the cadence of model training and deployment so that safety measures and regulatory oversight have time to catch up.

The plan, described in three stages, begins with a move Anthropic has already set in motion: opening its models to outside scrutiny. The company said it will grant independent evaluators such as METR broad access to its systems to verify that safety practices and public commitments are being honored.

The second stage, according to Amodei, would require collective action. Rather than any single lab unilaterally slowing down, the industry as a whole — likely with government involvement — would coordinate on shared pacing norms, reducing the competitive pressure that pushes companies to rush frontier models to market.

Amodei’s argument reflects a growing tension in the AI sector: labs racing to release ever-more-capable models while facing questions about cybersecurity risks, evaluation standards, and whether voluntary safety commitments carry enough weight. Anthropic’s decision to invite third-party auditing is positioned as a first step toward more formal accountability, and the company acknowledged it is acting on this step alone for now.

The debate over pacing extends well beyond the AI labs themselves. Companies across the technology landscape are watching how development norms evolve, since downstream businesses — from cloud providers to enterprise software makers — are building products on top of frontier models. Fintech firms are among the most active adopters of AI tooling. Block, Inc. (XYZ), which operates the Square and Cash App ecosystems, traded at $79.40 in recent action, down 1.0% from its previous close of $80.20, with a market capitalization of roughly $47.7 billion — a reminder that AI policy debates ripple through companies applying these technologies to payments and commerce.

Whether Anthropic’s call gains traction among rivals remains an open question. Coordinated slowdowns would be difficult to enforce in a competitive market with international participants, and the essay stops short of specifying enforcement mechanisms. Still, an explicit appeal for restraint from the leader of one of the field’s most prominent labs marks a notable shift in tone for an industry often defined by speed.

What to watch

• Details on the scope and timing of third-party evaluations by groups such as METR, and what their findings reveal.
• Responses from competing AI labs and whether any join or contest the pacing proposal.
• Policymakers’ reactions, including any movement toward coordinated AI development standards.
• Anthropic’s next frontier model release and whether its deployment timeline reflects the slower approach Amodei described.

Source: original release