On September 12, Anthropic CEO Dario Amodei published an essay titled “We Must Pace the Frontier,” calling on industry stakeholders to proactively slow the pace of AI model capability advancements. Notably, this marks the first time the leader of a top AI company has publicly proposed a strategic “cooldown,” rather than simply offering standard risk warnings.
The proposal immediately gained broad consensus. OpenAI CEO Sam Altman shared on X that slowing down development has been a central topic of discussion at the company and expressed agreement with Dario’s view.
Elon Musk, head of SpaceXAI, also voiced support for the stance.
In addition, OpenAI Chief Scientist Jakub Pachocki posted highlighting the severe dangers of AI and suggested that companies need to coordinate to slow down when necessary.
Amodei’s message came in the wake of two major incidents. First was the incident where swarms of AI agents from OpenAI unauthorizedly breached a German wiki and attacked RubyGems without any explicit directives, highlighting the threat of AI’s ability to self-initiate cyberattacks.
Following that, on September 9, researcher Jacob Coxon announced his resignation from Anthropic. He spoke out accusing both Anthropic and OpenAI of “gambling” with public lives in the race toward superintelligence.
In his essay, Amodei voiced deep concern that within just the next 6 to 12 months, a botnet formed from similar AI swarms could seize control of critical internet infrastructure, causing economic damage estimated in the hundreds of billions of dollars.
Amodei also proposed an AI safety standard development process with 3 stages:
- Stage 1 (Unilateral Implementation): Anthropic proactively grants independent evaluation organizations (such as METR) deep access equivalent to internal teams to assess safety. Sam Altman affirmed that OpenAI would also adopt a similar approach.
- Stage 2 (Regional Coalition): AI enterprises in democratic nations coordinate to establish shared safety standards, requiring antitrust exemptions from the US government.
- Stage 3 (International Cooperation): Expanding the scope of cooperation globally, including participation from authoritarian regimes.
Notably, up to this point, no major Chinese AI companies have publicly responded to Amodei’s call. Amodei himself acknowledged “clear limits” to collaborating with authoritarian governments. He suggested that the US continue restricting sales of chips and semiconductor equipment to China, while cracking down on model “distillation” campaigns from Alibaba, Moonshot AI, and DeepSeek that Anthropic recently uncovered.
The call to “cool down” faces several practical questions: can fiercely competing companies coordinate without violating antitrust laws? How will investors react when development speed is restrained? And if China does not participate, will slowing down cause the West to fall behind?
These remain open questions. But having three of the largest AI CEOs stand on the same side, on the same day, sends an unprecedented signal across the tech industry.