Anthropic CEO Dario Amodei Proposes Three-Step Plan to Deliberately Slow AI Capability Advances

Anthropic CEO Dario Amodei published an essay on Saturday proposing a three-step plan for AI companies to intentionally pace the rate at which they advance model capabilities, according to CNBC. The essay arrives amid what CNBC describes as a growing chorus of researchers calling for a coordinated slowdown in AI development.

The Three-Step Plan

According to Amodei, the plan is designed to temper development speed without "sacrificing commercial advantage or the United States' lead in AI," though he acknowledged some steps will be harder to implement than others.

Step one: Anthropic has "unilaterally" committed to granting third-party evaluators employee-level access to the company to verify safety practices and report incidents, Amodei said.

Step two: Leading AI companies in democratic countries should coordinate to establish common safety standards.

Step three: Democratic governments should coordinate with authoritarian governments on AI safety norms.

Amodei was explicit that this is not a call to halt progress. "To be clear, pacing does not mean halting model training or technical progress, but ensuring companies take adequate time to align and safeguard their models, and for third party evaluators to confirm this," he wrote.

Context: A Resignation and Renewed Extinction Warnings

The essay landed days after Jacob Coxon, a researcher who has worked at both Anthropic and OpenAI, publicly announced his resignation from Anthropic. Coxon said he quit because he believes Anthropic and OpenAI are "gambling with our lives," claiming that people building AI "earnestly believe it could kill us all by the end of the decade." The resignation triggered significant discussion on social media, according to CNBC.

Concerns about AI-driven catastrophic or extinction-level risk are not new. In 2023, a statement signed by prominent researchers and executives — including OpenAI CEO Sam Altman and Amodei himself — declared that "mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war."

Amodei addressed why he did not support a slowdown when that idea first circulated in 2023. At that time, he said, models were not powerful enough to act in the real world or capable of "significant deception, manipulation, cheating, or cyberattacks." His essay implies that calculus has since changed as models have grown more capable.

"I continue to believe that AI can enormously improve the quality of human life. My desire to achieve these benefits is undimmed," Amodei wrote. "But the benefits will only be achieved if we build the technology in the right way, and — so long as we use the time we gain well — it is worth taking unusually deliberate care to get it right."

The essay also arrives as Anthropic prepares for what is widely anticipated to be a major IPO, though the company has not disclosed an official timeline, according to CNBC.

What This Means

Amodei's proposal is notable less for its technical content than for its timing and framing: a leading AI lab CEO publicly advocating for slower capability advancement while his own company races toward a high-stakes IPO and continues frontier model development. The plan's first step — third-party access for safety verification — is concrete and unilateral, but steps two and three depend entirely on voluntary coordination among competitors and geopolitical rivals, a historically difficult proposition in any competitive industry, let alone one with national security implications. The essay reads as much as a response to internal and public pressure — following Coxon's resignation and broader extinction-risk debates — as it does a serious policy proposal. Whether rival labs like OpenAI, or governments in Beijing and Washington, engage with steps two and three will determine if this becomes more than a unilateral gesture from Anthropic.