Anthropic CEO Calls for Slower Growth in Frontier AI Capabilities
Dario Amodei wants safety measures and independent evaluation to keep pace with the growing capabilities of the most advanced AI systems. He is not calling for training or technological progress to stop.
Amodei’s call to slow AI is opening a debate over the pace of development of the most powerful models. Anthropic CEO Dario Amodei, in the essay “We Must Pace the Frontier,” is calling for a deliberate slowdown in the improvement of frontier AI models so that safety measures can keep pace with development.
Amodei is explicitly not calling for model training or technological progress to stop. His proposal is aimed at gaining time to align systems with safety measures and independently verify them. It is an intervention by the head of one of the leading AI companies in a debate involving safety concerns, economic competition, and geopolitical rivalry between the United States and China.
Amodei’s call to slow AI is built on three steps
Amodei proposed a framework with three basic elements:
- ongoing access for external evaluators to advanced systems,
- coordination among AI companies,
- broader international coordination.
Under the proposal, external evaluators would enable independent assessments of models and their safety measures. Coordination among companies would address the fact that the most powerful laboratories are also direct competitors. The third part of the framework extends the issue to the international level.
The proposal does not yet change any binding rules, and Anthropic has not announced an across-the-board slowdown in development. It has also not been confirmed that other major AI companies will adopt this framework or that its elements will become part of government regulations.
Industry support, political opposition
The AP independently confirmed that Sam Altman and Elon Musk had publicly expressed support for Amodei’s call. That support alone, however, does not yet represent specific joint commitments or the adoption of the same practices at OpenAI, Google, or xAI.
President Donald Trump took a different position. According to AP, he spoke out against slowing down if it could threaten the United States’ lead over China. The response shows a direct conflict between Amodei’s emphasis on the pace of safety efforts and the political emphasis on U.S. competitiveness.
Amodei also describes possible future risk scenarios in the text. His suggestion that AI could, within six to twelve months, control a swarm of agents capable of taking over the internet remains a prediction, not a confirmed event.
What will be important to watch
The next specific test will be whether Anthropic publishes details on how external evaluators would operate, including their mandate. It will also be important to see whether other companies announce measurable steps instead of general public support.
Attention will also focus on possible White House or congressional measures following the announced talks on AI. In his argument, Amodei also cites an incident involving OpenAI and Hugging Face, but its specific details and scope have not been verified in the available materials by a primary technical announcement from OpenAI.
Amodei’s call to slow AI therefore currently represents a proposal to change the approach to development, not an established rule for the entire industry.
Sources
- Dario Amodei — We Must Pace the Frontier – Primary text of Amodei’s proposal: slowing the growth of capabilities, independent evaluators, and industry and international coordination.
- Associated Press — Anthropic CEO Dario Amodei says AI industry needs to give safety measures time to catch up – Independently confirms the publication of the call, its safety rationale, and responses from other AI industry leaders.
- Associated Press — Trump downplays the need to check AI development – Confirms Donald Trump’s subsequent political response and the dispute over balancing safety guardrails with U.S. competitiveness against China.
Verified and updated: 09/14/2026 06:20
