A widening divide has opened among the world’s leading AI executives over how fast frontier AI development should proceed, with some pushing for a coordinated slowdown on safety grounds while others argue that any pause would hand a decisive advantage to global rivals.
Anthropic CEO Dario Amodei has proposed a three-step plan for pacing AI development, calling for stronger safety testing and independent evaluation of advanced models, closer coordination among leading AI companies, and broader international cooperation. OpenAI CEO Sam Altman responded by reposting Amodei’s proposal, saying he agreed on the need to pace the frontier and that the idea had already been a major topic of internal discussion at OpenAI, adding that the company intends to adopt Anthropic’s proposal by giving independent evaluators employee-like access to its systems. Altman also said the public was right to be concerned about AI’s risks, though he expressed some reservations about how the slowdown argument has been framed, warning that conditional commitments, where companies act responsibly only if rivals do the same, risk deepening public anxiety rather than easing it. He pointed instead to the need for greater rigour around cybersecurity, citing recent instances of AI systems being used in attacks on other companies.
Elon Musk has also backed calls for a coordinated slowdown, having previously supported a 2023 push by the Future of Life Institute for a six-month pause in AI development to allow time for stronger safety measures.
On the other side of the debate, Nvidia CEO Jensen Huang has argued against decelerating development or introducing new regulation, reflecting a broader camp within the industry that views a slowdown as strategically risky given intensifying competition from China. US President Donald Trump weighed in as well, writing on social media that AI needs no guardrails beyond a strong, capable president, a remark that appeared to dismiss the substance of the safety debate altogether.
Governments have offered a more cautious response than parts of the industry. Germany said it was closely monitoring global AI developments and taking public warnings from leading developers seriously, while officials elsewhere have floated the idea of an international agreement to manage AI risks, drawing comparisons to existing nuclear non-proliferation frameworks.
The disagreement comes as both OpenAI and Anthropic are reportedly preparing for what could be record-breaking public listings, with Altman confirming plans to delay OpenAI’s IPO until next year in order to prioritise safety work, even as Amodei has acknowledged that the industry’s overall pace of advancement has surprised even those building it.
