The heads of four companies that have fiercely competed in artificial intelligence (AI) development—Anthropic, OpenAI, SpaceX's xAI, and Google DeepMind—have united in a rare consensus, agreeing that "the speed of AI development must be slowed down."
In an interview with CBS News broadcast on the 13th (local time), Dario Amodei, CEO of Anthropic, compared the AI advancement process to a curve progressing as "1, 2, 4, 8, 16, 32," diagnosing that it is currently at the stage where the curve "is beginning to steepen."
"We don't need to panic today. We don't need to shut everything down," he said, adding, "Nevertheless, this is a warning sign that we need to slow down."
In the interview, Amodei also stated, "I think for a long time the industry has lied to people about the risks of this technology."
He further noted, "U.S. companies should not sell high-performance AI chips to China," adding, "The export of these chips will be a key factor determining China's AI capabilities."
Acknowledging that the "toughest dilemma" regarding his pacing proposal is the possibility that hostile nations like China might not take similar measures, he maintained his stance.
While citing arms reduction agreements with the former Soviet Union as a potential model, he added, "Because the motive to get ahead and the military advantage of doing so are so great, it is going to be very difficult, and frankly I'm not sure if it's possible, but we have to try."
In a post published on his personal blog the previous day, Amodei presented two incidents supporting his argument to slow down AI development.
One is the growing ability of AI systems to perform "recursive self-improvement," in which they create faster versions of themselves without human intervention, and the other is an incident last July where a swarm of OpenAI's AI agents broke out of a testing environment to hack Hugging Face, an open-source AI model sharing platform.
He warned that if a more powerful swarm of agents "were to cause a similar level of alignment failure, it could lead to catastrophic harm," and cautioned that such rogue agent swarms could take over the internet within six to 12 months.
He then proposed a three-stage speed-regulation plan that includes placing third-party safety evaluators inside AI companies, establishing common standards among democratic nations, and cooperating with authoritarian countries.
OpenAI CEO Sam Altman immediately echoed Amodei's arguments.
Writing on X (formerly Twitter), Altman said, "As Dario said, I agree that we should moderate the speed of the frontier. Having independent evaluators with employee-level access is a great idea, and we will take the same steps."
In a previously released interview with Fortune, Altman stated, "I think taking a risk that by the end of this decade there's a 10% chance of killing all humans is an unacceptable trade-off."
He also stated, "Given everything going on with safety, now is not the right time to go public, and we don't feel any such pressure," hinting that OpenAI is unlikely to hold an initial public offering within the year.
SpaceX CEO Elon Musk also shared Amodei's post on X, briefly agreeing by writing, "Dario is right."
Musk, who is also a co-founder of OpenAI, later merged his AI company, xAI, into SpaceX.
Google DeepMind CEO Demis Hassabis also posted on X, stating, "Dario's essay points in the right direction. The details need to be worked out, but it's the right direction for such a critical moment."
Amodei's remarks came just days after Anthropic researcher Jacob Coxon resigned on the 9th, warning that AI could threaten humanity and stating that he did not want to be part of an industry perceived to be in a "state of out-of-control."
With Musk, Altman, and Hassabis successively aligning with Amodei's stance, a rare scene unfolded in which the heads of four companies that usually compete in cutting-edge AI development stepped in lockstep within a single day.
(Photo: AP, Yonhap News)
※
Copying, redistribution, and unauthorized use in AI training are strictly prohibited.