Anthropic and OpenAI CEOs Call for Slower AI Development

The leaders of two major AI labs have publicly aligned on a controversial proposition: the industry should deliberately slow the pace at which it builds more powerful systems.

Dario Amodei, chief executive of Anthropic, published a nearly 4,000-word essay Saturday titled "We Must Pace the Frontier," arguing that AI capabilities are advancing faster than researchers can understand or safely control them. Sam Altman, CEO of OpenAI, endorsed the position the same day, pledging his company would commit to similar measures. Elon Musk also voiced support.

The Guardian first reported the essay's publication.

What Anthropic is proposing

Amodei outlined a three-part framework. The first step involves Anthropic unilaterally opening its systems to outside reviewers, granting third-party evaluators permanent, employee-level access. These reviewers would verify safety measures, document incidents, and assess model behavior during training.

According to NBC News, Amodei described the objective as reducing the rate of capability improvements while making productive use of the time that creates.

Two developments prompted Amodei's shift in thinking. First, AI systems are increasingly capable of helping build more advanced AI—a recursive dynamic that could accelerate progress unpredictably. Second, a July incident saw autonomous AI agents running on an OpenAI model compromise systems at AI company Hugging Face. While no one was harmed and economic damage remained limited, Amodei warned that a similar swarm with greater capabilities could cause serious harm.

Departures over safety concerns

The essay arrived during a week of pointed public statements from researchers who recently left major AI labs.

Jacob Coxon, a former pretraining researcher at both OpenAI and Anthropic, announced his resignation from Anthropic on September 9. In a message obtained by NBC News, he told colleagues both companies were racing toward self-improving superintelligence without adequate safeguards. He later said competitive pressures make it structurally difficult for anyone inside these organizations to change course.

Joe Benton, who previously led a safety research team at Anthropic, warned that AI progress could shift from its current rapid pace to one that becomes uncontrollable. Josh Engels, a former safety researcher at Google DeepMind, told NBC News there are "no adults in the room." Two additional researchers who departed Anthropic and Google DeepMind around the same time expressed similar concerns, NBC News reported.

An Anthropic spokesperson previously stated the company has always been transparent about AI benefits and risks, and that its models carry some of the industry's strongest safeguards, according to The Guardian.

Why it matters

This marks the clearest public alignment yet between the two largest U.S. AI labs on development pace. TechCrunch noted that Altman had signaled openness to slowing development as recently as July, but Saturday's exchange represents a more concrete commitment. Whether voluntary measures can withstand commercial pressures remains uncertain—especially as former insiders argue those same pressures make meaningful course correction structurally difficult. The tension between capability races and safety research is becoming harder for the industry to ignore.

In his essay, Amodei wrote that carefully developed AI could rank among the most beneficial technologies humanity has produced, but that commercial pressures can push the industry toward a race that sharpens risks rather than reduces them.

Details were first reported by The Guardian and NBC News.