Dario Amodei, CEO of Anthropic, published an essay over the weekend calling on the AI industry to slow the pace of developing the most advanced models. He set out a three-point plan that he argues would decelerate development while preserving market competitiveness and the United States' leadership in AI.
The proposal in brief:
- An Anthropic unilateral commitment to give external evaluators employee-level access to inspect safety practices and report incidents.
- Collaboration among leading AI companies in democratic countries to establish common safety standards.
- Coordination between democratic and authoritarian governments to reduce global risks.
The proposal drew endorsements from several prominent figures: Sam Altman, CEO of OpenAI; Demis Hassabis, President of Google DeepMind; and Elon Musk. Altman said his company is unlikely to go public this year, partly for safety reasons, and posted on X that he agrees the frontier needs to be paced. Hassabis described the essay as pointing in the “right direction,” while Musk signaled support for cross-company safety reviews.
The central dilemma: international cooperation
In a Sunday interview with CBS, Amodei acknowledged that his plan’s toughest problem is what happens if rival powers—chiefly China—do not join a slowdown. He highlighted the strong incentives for gaining advantage and the potential military benefits, saying he was unsure whether global cooperation is possible but that it must be attempted.
The international dimension is central because technological leadership carries geopolitical consequences.
Pushback and skepticism
The essay also prompted critical reactions. Jacob Coxon, an Anthropic researcher, recently resigned, saying the company and OpenAI were “playing with our lives.” David Sacks, a former AI and cryptocurrency official in the Trump administration, voiced skepticism: if the firms truly believe their models are dangerous enough to slow down, they should do so, he argued, but not frame the action as requiring others’ permission. Sacks added that responsible development can also be sound business, given the product-liability risks.
Alexandr Wang, head of Meta’s AI division, did not respond directly to the essay but indicated that Meta Superintelligence Labs is allocating more resources to alignment—making AI systems conform to human values.
U.S. political reactions
Donald Trump sought to downplay some of the alarm about AI, telling reporters (via Bloomberg) that the United States must retain its edge over China. He said the U.S. is the world’s most technologically advanced country and that he wants to keep that position because “whoever wins in AI wins everything,” adding that he believes only “negative forces” are inflating the risks.
Kevin Hassett, Director of the National Economic Council, called making AI safe a “manageable problem” on Fox News and welcomed Amodei’s proposal as a potential model for the private sector. He noted that Sean Cairncross and Michael Kratsios in the administration are also examining the issue.
Timing: amid regulatory, social and market pressure
The call to slow development comes at a critical moment: AI companies face regulatory scrutiny and growing public unease while Anthropic and OpenAI are preparing for potentially historic initial public offerings. Public resistance has focused on local strains from new data centers, rising electricity bills and fears of job losses—issues that have become part of the November midterm campaign conversation.
Markets are also under pressure as investors debate whether massive investments in AI and supporting infrastructure will pay off; high-priced tech stocks have proved especially sensitive to increased spending or weaker returns.
Conclusion
Amodei’s three-point plan has attracted significant industry support, but its success depends on whether global cooperation—especially including China—can be achieved. The debate spans technical, business and geopolitical dimensions as companies try to balance safety concerns, commercial incentives and international risk.
Publication notes
Press coverage of the essay and ensuing reactions appeared in mid-September 2026. This article was prepared with assistance from an AI tool; the final content was edited and verified by our journalist.
