(The Epoch Times)—Anthropic CEO Dario Amodei is calling on artificial intelligence companies to slow the development of their most advanced models, a proposal that has drawn support from several other industry leaders.
In a lengthy essay published Sept. 12, Amodei said that simply investing in risk prevention is no longer enough. Instead, he said companies developing frontier models should deliberately slow the rate at which they increase their capabilities.
Doing so, he said, could give researchers and governments more time to develop safeguards before AI systems become too powerful.
“I believe that if slowing down bought us even an extra year or two before models reach critical levels of capability, and we used that time to advance alignment, we could greatly reduce the risk that something goes seriously wrong,” Amodei wrote.
“Progress will still seem fast, and we must make wise use of the time we gain,” he added.
One of his main concerns is AI’s growing ability to help build the next generation of AI systems.
That process, often called recursive self-improvement, has accelerated since roughly the summer, Amodei said. He warned that AI-assisted research could eventually outrun human capabilities to manage AI systems.
“Left unchecked, it could outrun our ability to understand and control these systems,” he wrote.
Industry Leaders Back Proposal
In a rare moment of unity, leaders and researchers at rival AI companies rallied around Amodei’s call.
“I agree with Dario that we need to pace the frontier,” OpenAI CEO Sam Altman wrote on X, adding that OpenAI would match Anthropic’s commitment to give independent evaluators employee-like access to its systems.
Elon Musk, CEO of SpaceXAI, also backed Amodei on Sept. 12, writing on X: “Dario is right.”
Hugging Face CEO Clement Delangue said his company was launching an “open alignment initiative” and asked to participate in efforts to give independent evaluators deeper access to AI labs.
Andrej Karpathy, an AI researcher and former OpenAI and Tesla executive, also endorsed the proposal.
“I love this and really hope we can come together as an industry and make it happen,” Karpathy wrote on X.
Google DeepMind CEO Demis Hassabis had not publicly responded to Amodei’s proposal by the time of this publication. He has, however, previously called for greater outside scrutiny of frontier AI.
A 3-Step Plan
Amodei proposed a three-stage framework for “pacing the frontier.”
The first would give independent safety evaluators permanent, employee-like access to frontier AI companies. Anthropic said it is willing to adopt that measure itself and asked other major developers to follow.
The second step would require frontier AI companies to coordinate on common safety standards so that one company is not placed at a competitive disadvantage for slowing down.
The third—and arguably most difficult—would involve making agreements with China to prevent an AI race in which developers feel forced to advance faster than safety measures can keep up.
Amodei said governments would likely need to become involved, both to establish enforceable standards and to address legal concerns surrounding cooperation among competing companies.
China Complicates Any Slowdown
Any industry-wide slowdown would face a major geopolitical obstacle: the U.S.-China tech race.
Amodei argued that U.S. companies cannot simply slow indefinitely if Chinese developers continue advancing at full speed.
“Pacing within democracies will be limited by the lead that U.S. companies have over authoritarian regimes, chiefly the Chinese Communist Party,” he wrote.
Amodei called for tighter restrictions on advanced AI chips and measures to prevent the theft or copying of leading U.S. models. At the same time, he said that Washington should eventually seek agreements with Beijing on at least some AI safety issues.
Even limited cooperation, he said, could give both sides more room to slow development when dangerous capabilities emerge.
The proposal attracted the attention of members of Congress.
Sen. Brian Schatz (D-Hawaii) wrote on X that he is “studying” Amodei’s proposal, which he called “a reasonable start” that “takes seriously” the need for measures that can be enacted quickly.
Sen. John Cornyn (R-Texas), meanwhile, expressed skepticism on China’s willingness to cooperate. “Will China?” he wrote.
The prospect of an AI slowdown could emerge as a topic when President Donald Trump meets China’s Xi Jinping in Washington later this month. Trump has confirmed that artificial intelligence will be among the issues the two discuss.
“[Chinese leader] Xi is coming over on September 24th, and we talked about [artificial intelligence] when I was over in Beijing, and we’ll be talking about it again,” Trump said in July.
The United States and China are also preparing for a broader dialogue on AI, expected to address issues including cybersecurity, advanced models, and the risks posed by increasingly autonomous systems.
For Amodei, international competition does not eliminate the need to slow down.
Instead, he argued, it makes coordination increasingly important.
“The benefits will only be achieved if we build the technology in the right way,” he wrote.
