Dario Amodei, co-founder and CEO of Anthropic, speaks to “The Circuit with Emily Chang” at Anthropic’s headquarters in San Francisco, California, on Thursday, April 30, 2026.
Jason Henry | Bloomberg | Getty Images
Anthropic CEO Dario Amodei published an essay on Saturday urging artificial intelligence companies to quickly improve the capabilities of their models. The move comes as researchers increasingly call for a coordinated slowdown.
Amodei proposed a three-step plan to help slow the pace of development without “sacrificing commercial advantage or America’s lead in AI,” although he acknowledged that some steps may be easier to achieve than others. Anthropic is actively preparing for what is widely expected to be a historic IPO, but has not officially announced when it plans to debut.
Amodei said Anthropic is “unilaterally” working on the first phase of the plan, which will give third-party evaluators employee-level access to the company to examine safety practices and report incidents. The second step would encourage major AI companies within democracies to coordinate and establish common safety standards, and the third step would require coordination between democratic and authoritarian governments.
“To be clear, pacing does not mean stopping model training or technological advances, but rather ensuring that companies have sufficient time to adjust and secure their models, and that third-party evaluators review this,” Amodei wrote.
Concerns about AI capabilities
Amodei’s essay was published after a human researcher sparked a firestorm on social media this week by announcing he was leaving the company. Jacob Coxon, who also worked as a researcher at Anthropic’s main rival, OpenAI, said he resigned over concerns that Anthropic and OpenAI were “putting our lives on the line.” He said those developing AI “seriously believe that AI could kill us all by the end of the decade.”
Although extreme, concerns about the potential for AI to cause human extinction or other catastrophic events are not new in AI research circles. For example, in 2023, prominent AI researchers and executives, including Amodei and OpenAI CEO Sam Altman, signed a statement stating that “reducing the risk of AI extinction should be a global priority, alongside other society-wide risks such as pandemics and nuclear war.”
Amodei said on Saturday that there was talk of a pause or slowdown in AI development starting in 2023, but that it “made little sense” to do so at the time. He said that at that point, the model was not powerful enough to take action in the real world, nor was it capable of “significant deception, manipulation, fraud, or cyber-attacks” yet.
“I continue to believe that AI can significantly improve the quality of human life. My desire to achieve these benefits remains undiminished,” Amodei wrote. “But that benefit is only achieved if you build the technology the right way. As long as you make good use of the time you get, it’s worth taking more careful care than usual to get it right.”
Support for voluntary deceleration
Mr. Amodei’s essay was praised Saturday by many industry researchers and executives, including Mr. Altman. In a post about X, he said he agreed with Amodei that the industry needs to accelerate the development of advanced AI capabilities. Altman said the topic has been a “major topic” of discussion at OpenAI in recent weeks.
“It’s a great idea to work on having independent evaluators with the same access rights as employees, and we intend to do the same,” Altman said. “We have more to share soon.”
Jakub Pachocki, chief scientist at OpenAI, published a blog post earlier this month warning that no AI company has “solved tuning and monitoring well enough to continue to scale responsibly at full speed over time.” In the AI industry, coordination refers to the work done by AI developers to ensure that systems behave in accordance with human values and intentions.
Paciocchi said he hopes and expects that “voluntary slowdowns will become the norm until a shared safety bar is established.”
Elon Musk also expressed support for the economic slowdown in a post on X on Saturday, writing, “Dalio is right.”
Musk’s competing AI startup xAI has been acquired by his own rocket company space x Earlier this year, he was a vocal critic of Anthropic. He previously said the company “hates Western civilization” and is “destined to be the opposite of its name,” but that would be misanthropy. But since Anthropic announced a major computing deal with SpaceX in May, Musk has changed his tune significantly.
“Everyone I met was extremely competent and extremely concerned about doing the right thing,” Musk wrote at the time. “No one set off my evil detector.”
Amodei wrote on Saturday that he believes AI still has the potential to “drastically improve the quality of human life,” but that the risks need to be taken seriously.
“We believe that if we can take even an extra year or two to slow down and allow the model to reach a critical level of performance, and then use that time to make further adjustments, we can significantly reduce the risk of something major going wrong,” he said.
Attention: Human AI researcher says company is ‘risking our lives’
