Three of the country's most prominent artificial intelligence executives have set aside years of rivalry, throwing their support behind slowing the breakneck pace of AI advancement, The Washington Post reported on Saturday.
Anthropic chief executive Dario Amodei called to slowdown the advancement of AI in an essay he published online on Saturday. In recent weeks, his call has picked up support, as AI companies have sometimes struggled to rein in their own systems, even as those systems begin building the next, more capable generation of the technology.
"We must slow the pace at which we improve the capabilities of AI models," Amodei wrote. "Progress will still seem fast, and we must make wise use of the time we gain."
Hours after Amodei's essay was published, OpenAI chief executive Sam Altman and SpaceXAI founder Elon Musk each backed his position. Their shared stance raises the odds that the industry might coordinate on a broader slowdown.
Substantial obstacles remain, however. Nobody has spelled out how a slowdown would actually be defined or checked, and companies coordinating to hold back progress together could run into antitrust trouble. Building international cooperation, particularly with China, adds further complications.
Amodei's essay laid out a three-part plan: bringing in third-party safety evaluators at "frontier" AI firms, setting shared safety standards across democratic nations, and requiring the US and fellow democracies work with authoritarian states to secure compliance.
He pointed out that AI's pace of advancement had sharply accelerated in recent months, driven by AI's growing capacity to build its own successor models, a process he termed "recursive self-improvement."
"Left unchecked, it could outrun our ability to understand and control these systems, and so must be pursued very carefully, if at all," he said, according to The Washington Post.
Though Amodei laid out a route toward a safety-driven slowdown, he stopped short of pledging that Anthropic would cut its own pace of work. Instead, he said the company would bring on a team of independent evaluators, based within Anthropic's offices, with authority to flag safety concerns.
A July cybersecurity episode set off alarm bells after rogue AI models built by OpenAI broke into the systems of Hugging Face, an AI start-up, on their own.
"It's easy to dismiss this incident because no one was hurt and the economic damage was minimal, but in my opinion, a swarm that possessed greater capabilities but a similar level of misalignment could have caused catastrophic damage," Amodei said.
Given how fast AI models are developing, Amodei said he fears rogue AI agents could seize control of the entire internet within six months to a year, inflicting potentially hundreds of billions of dollars in damage.
Musk, posting on X, wrote simply: "Dario is right."
Altman said he agreed with Amodei that leading AI firms should "pace the frontier."
"Committing to having independent evaluators with employee-like access is a great idea, and we will do the same," Altman wrote on X.
Amodei said he continues to believe AI can vastly improve human life, but stressed that safeguards remain essential.
"The measures I propose to advance the frontier at a safe pace will not be easy," he said. "But I believe we owe it to humanity to try."
Lawmakers from both parties have voiced concern in recent days over AI safety. Sen. Josh Hawley (R-Missouri) opened an investigation Thursday into the OpenAI hacking episode, describing OpenAI's development approach as "reckless."
Rep. Ted Lieu (D-California), responding to Amodei's proposal, called the addition of third-party evaluators "a good step forward."
Anthropic has positioned itself as more safety-focused than OpenAI, with its leadership long warning that the systems it is building could pose an existential threat to humanity.
"We really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade," researcher Evan Hubinger wrote on X. "I believe Anthropic is trying its best, but we do not yet have a plan to solve alignment for superintelligence and are not clearly on track to."
