Amodei, Altman and Musk Find Common Ground for Once: “AI Development Must Be Slowed”

[Financial News]

The CEOs of leading AI companies—including Dario Amodei of Anthropic, Sam Altman of OpenAI and Elon Musk of SpaceX—have found common ground for once. They jointly called for slowing AI development. As concerns grow that AI could trigger a catastrophe resulting in humanity’s extinction, United States President Donald Trump is pushing ahead with development, saying, "We cannot afford to fall behind China." Yet the industry’s leading figures are taking a different position.
According to the Financial Times (FT), Altman and Musk strongly endorsed the proposal after Amodei made the suggestion in an essay published on the 12th (local time). Musk said on X, "Dario is right."
In an essay published earlier on his blog, Amodei said the pace of AI model development this summer had been "intensely rapid" and argued, "We need to slow the pace at which AI models are improving." Although Anthropic is preparing the largest initial public offering in its history, its CEO has raised the potentially damaging issue of limiting development speed.
Altman and Musk, who had mostly traded barbs, responded to Amodei’s appeal for once. They jointly supported his argument and agreed that, as Amodei proposed, an independent evaluation group should be given access to all frontier AI companies.
Musk said, "Dario is right," while Altman emphasized, "I agree with Dario that we need to regulate the pace of frontier development." Earlier, Altman had also said at an all-hands meeting at the beginning of last week that he could accept international cooperation to slow AI development.
In a post on X, Altman said, "It is a great idea to allow independent evaluators to have access as if they were internal employees," adding, "We will do that too. We will share more soon."
Executives at Google DeepMind and Meta Platforms have not yet stated their positions. However, concerns are already growing within those organizations. Demis Hassabis, president and co-founder of Google DeepMind, said in an essay last July that rapid AI progress requires new approaches to inspecting frontier systems.
Concerns that rapid AI development could spiral out of control spread quickly last week after Jacob Coxon, a researcher at Anthropic, resigned while raising concerns about AI safety and human extinction. In his resignation on the 8th, he criticized Anthropic and its competitors, saying, "They are gambling with our lives." He said the companies were focused solely on competing to develop superintelligent AI that could improve itself so rapidly that humans would be unable to maintain oversight.
Coxon said, "Those developing AI genuinely believe that AI could kill all of us by 2030."
His warning came especially after it emerged that a group of OpenAI agents had escaped a test system and hacked Hugging Face, an open-source AI platform. The OpenAI agents reportedly caused the incident while carrying out tasks assigned by developers: They took control of OpenAI headquarters’ servers after an error and acted through cooperation among the agents themselves.
Amodei said, "If these agents had operated with more powerful capabilities, they could have caused catastrophic damage," adding that the incident had become a decisive motivation for his proposal to slow AI development.
He added, "Similar incidents, although less serious than this one, have occurred across the industry, including at Anthropic," and said, "Every frontier AI company has a responsibility to treat the Hugging Face incident as if it had happened at its own company and respond accordingly."
Amodei urged government support, judging that industry coordination to regulate the pace of development could be viewed as collusion and violate antitrust laws. He also warned that Chinese AI companies are growing rapidly and called for cooperation with the Chinese government to strengthen AI controls.
He emphasized that this could also address President Trump’s concern that "the United States would be placed at a severe disadvantage if it failed to win the AI race."
Amodei said, "Even with a coordinated strategy to regulate the pace, frontier AI developers could gain time to carry out this critical safety work without sacrificing commercial advantages or the United States’ leadership in AI."

[email protected] Song Kyung-jae Reporter