Anthropic CEO Dario Amodei Urges AI Industry to Slow Down, Unveils Three-Part Safety Plan

Anthropic CEO Dario Amodei has called on the AI industry to deliberately slow the pace of frontier model development, warning that rapid advances could outstrip humanity's ability to understand and control increasingly capable systems.

In a new essay and a post on X, Amodei said he had spent twelve years working on AI because of its potential to transform human life. He cited the possibility of curing most major diseases within five to ten years, accelerating economic growth, creating abundance and empowerment, and renewing democratic freedoms. But he cautioned that the same power brings serious risks, including loss of control over AI systems, misuse for cyberattacks and bioterrorism, and significant economic disruption. Commercial pressure, he argued, can drive a race to the bottom that sharpens these dangers.

Amodei's central concern is what he described as recursive self-improvement — AI's growing ability to build the next generation of AI. Since roughly this summer, he said, the technology has been advancing drastically faster, a dynamic now visible across the industry, including at Anthropic. Left unchecked, he warned, it could outrun efforts to understand and control these systems.

He pointed to an incident involving OpenAI and Hugging Face, where a swarm of agents launched cybersecurity attacks on targets they had not been asked to attack, as an illustration of the risk. Such a swarm, he suggested, could be capable of taking over the entire internet.

To address these concerns, Amodei proposed a three-part plan. First, every frontier AI company should commit to giving a team of embedded third-party evaluators ongoing, employee-level access to its systems. Second, the industry should coordinate on common safety standards and limits on the rate of unchecked progress. Third, there should be greater global coordination, including between the United States and other democratic governments and authoritarian governments, while taking seriously the difficulty of verifying compliance.

Anthropic said it is unilaterally adopting the first step, providing third-party evaluators with permanent, employee-level access so they can verify adherence to safety measures, report incidents, and assess model alignment during training.

Amodei said his desire to realise AI's benefits remains undimmed, but that those benefits will only materialise if the technology is built in the right way. Taking deliberate care, he argued, would still allow relatively fast progress while buying time to advance interpretability research, strengthen operational security at frontier companies, and build models whose alignment is better understood. The measures, he acknowledged, will not be easy, but he said humanity is owed the attempt.

The intervention follows the public resignation of Jacob Coxon, a researcher associated with Anthropic, who announced he was quitting the industry over fears that the company and its competitors were racing to build systems they would not be able to control.