Anthropic CEO Dario Amodei is calling for frontier AI development to slow so safety research, external evaluation and regulation can keep pace with increasingly capable models. OpenAI CEO Sam Altman and xAI founder Elon Musk have publicly backed parts of the proposal, while U.S. President Donald Trump has rejected slowing development if it risks giving China an advantage.
In his essay We Must Pace the Frontier, Amodei said he does not support stopping AI development. Instead, he wants companies to spend more time aligning and safeguarding powerful models before advancing their capabilities further.
Amodei Proposes Three Levels of Oversight
The first part of Amodei’s plan would give independent evaluators ongoing access to frontier AI companies, allowing them to examine safety practices, training processes and emerging models. Anthropic said it would commit to this approach itself and wants governments to require comparable measures from other major developers.
His second step calls for coordination across the AI industry, while the third would require international cooperation. Amodei said even an additional year or two before models reach critical capability levels could give researchers more time to improve alignment, interpretability, security and operational safeguards.
Amodei pointed to rapid progress in AI systems that can help develop their successors, as well as recent safety incidents involving autonomous agents. One example was the OpenAI-Hugging Face cybersecurity incident, in which agents attacked systems outside their assigned targets and attempted to interfere with the system evaluating them.
He said a more capable system behaving similarly could cause far greater damage. Amodei estimated that within six to 12 months, a comparable group of agents with stronger capabilities could potentially establish a persistent botnet across large parts of the internet.
AI Leaders Agree on More Safety Checks
Altman said he agreed that frontier development should be paced and described independent evaluators as a strong proposal. Musk also backed Amodei’s position, while Google DeepMind CEO Demis Hassabis has expressed support for the general direction.
The calls follow warnings from former Anthropic researcher Jacob Coxon, who said some employees at AI companies are seriously concerned about the consequences of rapid development. Coxon argued that any meaningful slowdown would also require coordination with China to prevent safety measures in the U.S. from simply becoming another international race.
Amodei acknowledged the same difficulty. He argued that any slowdown would need to preserve the U.S. competitive position and said cooperation with China could increase the time available for safety work.
Trump has taken a different position, saying the U.S. must maintain its AI lead over China. Critics have also questioned whether proposals from major AI companies could strengthen the position of established developers by making it harder for smaller or open-source competitors to operate.
Featured image credits: Wikimedia Commons
For more stories like it, click the +Follow button at the top of this page to follow us.