Anthropic CEO Dario Amodei has advocated for a deliberate slowdown in AI frontier model advancement to allow time for safety research, echoing OpenAI's Sam Altman. The push comes as AI companies invest heavily in building more capable systems, raising concerns about insufficient safeguards against emerging risks.
- Amodei calls for slowing AI to align with safety research
- Proposes third-party evaluators inside AI companies
- Calls for international coordination and export controls
What happened
Days after OpenAI CEO Sam Altman expressed openness to slowing AI development, Anthropic CEO Dario Amodei published an essay titled “We Must Pace the Frontier,” urging the AI industry to decelerate progress in frontier models. Amodei highlighted that rapid AI capability growth risks outstripping safety controls and testing capacity, potentially leading to dangerous exposures.
Amodei emphasized that pacing does not mean stopping AI research but rather limiting capability improvements enough to allow companies, independent evaluators, and governments to keep safety measures up to date. Prominent figures like Sam Altman and Elon Musk publicly supported these calls, proposing collaborative frameworks to better oversee AI system development.
Why it matters
The AI industry currently faces heightened competition to build increasingly powerful models, investing billions in computing infrastructure. This race risks pushing companies to deploy insufficiently tested systems, potentially leading to severe cybersecurity threats and other adverse outcomes. Amodei specifically cited concerns about recursive self-improvement and agentic AI systems that could autonomously create botnets or take over internet components within the next year.
By advocating for pacing, Amodei aims to ensure safety research—including improved operational controls, alignment techniques, interpretability studies, and robust evaluation methods—can progress alongside AI capabilities. This approach seeks to prevent technology outstripping regulatory and safety frameworks, which could exacerbate risks for societies globally.
What to watch next
Amodei proposed a three-pronged strategy involving embedding independent external evaluators into AI companies with employee-like access, encouraging coordination among frontier AI firms in democratic countries, and fostering global cooperation backed by government support. He also emphasized stricter export controls on advanced AI chips and equipment, particularly to China, to maintain leadership and safeguard against unauthorized model usage.
Going forward, stakeholders will monitor how AI companies implement these safeguards, whether independent evaluators gain significant access, and how international coordination evolves. The unfolding regulatory responses will shape whether such pacing efforts can effectively balance innovation with mitigating AI-driven risks, especially as new frontier models continue to emerge.