Anthropic CEO Outlines Plan to Slow AI Development

Anthropic CEO Dario Amodei has outlined a plan to slow the pace of AI development, proposing that independent third-party evaluators such as METR be granted access to its models to verify adherence to safety practices and commitments. The proposal appeared in a lengthy essay in which Amodei described a three-step approach to pacing the frontier, a phrase that has become shorthand for coordinated restraint among leading AI labs. The core idea is that voluntary commitments are not enough on their own. Without external verification, labs can claim to follow safety practices without anyone able to confirm it. By inviting outside evaluators in, Amodei is arguing that credibility requires transparency and independent scrutiny, even when that scrutiny is inconvenient. The three-step framework is intended to give labs a structured way to slow down when certain risk thresholds are approached, rather than relying on ad hoc decisions made under competitive pressure. That pressure is intense: the race to build more capable models has accelerated dramatically, and any lab that unilaterally pauses risks falling behind rivals who do not. What makes the essay notable is the apparent alignment between competitors. OpenAI's Sam Altman has voiced similar sentiments about the need for coordinated safety measures, suggesting an unusual degree of convergence between two of the leading AI companies. Whether that convergence reflects genuine shared concern or a strategic effort to shape regulation remains an open question. Skeptics point out that voluntary pacing faces a fundamental collective action problem. If only some labs slow down, others may capture the benefits of moving faster. Enforcement, verification, and international coordination all remain unresolved. Still, the proposal adds momentum to a broader conversation about how frontier AI development should be governed, and it puts pressure on other labs to explain how they would verify their own safety claims.