Anthropic CEO Proposes Three-Stage Plan to Slow Frontier AI Development

Anthropic CEO Dario Amodei has called for a deliberate slowdown in the development of frontier AI capabilities to ensure societal preparation and safety. In a blog post published September 12, 2026, Amodei outlined a three-stage strategy to "pace the frontier," arguing that the benefits of AI will only be realized if the technology is built with unusually deliberate care.

As an immediate first step, Anthropic is unilaterally committing to a model of "embedded evaluators." The company will grant third-party organizations, such as METR, deep access to its internal operations. These external evaluators will be provided with company badges, desks, and laptops, granting them access comparable to that of Anthropic's own internal risk assessment teams. Amodei cited two primary catalysts for this shift: the OpenAI-HuggingFace hack and the increasing ability of AI systems to assist in building the next generation of AI models.

A Tiered Approach to Coordination

The second stage of Amodei's proposal focuses on industry-wide coordination among AI companies within democratic nations. He suggests establishing common safety standards and limits on the rate of unchecked progress. To navigate the legal complexities of such an agreement, Amodei proposes that the U.S. government mediate these discussions to avoid potential antitrust violations.

The final stage envisions global coordination that extends to authoritarian governments, including China. This international cooperation would specifically target the prohibition of high-risk applications, such as the use of AI for the production of biological weapons.

The Pressure for a Brake Pedal

This proposal arrives during a period of escalating tension within the AI safety community. The call for pacing follows a July 2026 letter signed by more than 1,100 AI insiders urging the U.S. government to develop tools for a potential industry slowdown. The urgency is further highlighted by the recent resignation of Anthropic researcher Jacob Coxon, who warned that leading AI firms are "gambling with our lives."

Despite the call for a slowdown, Amodei believes the U.S. can maintain its competitive edge. He argues that by cracking down on model distillation and restricting the sale of critical chips, the U.S. could significantly widen its lead over China over the next three to five years, even while pacing the development of new capabilities.

Implications for the Industry

Amodei's plan represents a shift from theoretical safety warnings to a concrete operational framework. By opening its doors to embedded evaluators, Anthropic is attempting to establish a new precedent for corporate transparency and regulatory trust. The move is a direct attempt to create an industry "brake pedal," aiming to prevent a race to the bottom where safety is sacrificed for speed.

What remains to be seen is whether other frontier labs will follow Anthropic's lead in granting deep internal access to third parties, or if the proposed government-mediated coordination can overcome the competitive pressures of the AI arms race.