Who gets to decide how quickly AI moves?

Senior executives at several leading U.S. AI firms have proposed slowing the rollout of the most advanced models, arguing that existing safety measures cannot keep pace with rapid capability gains. The call was initiated by Anthropic CEO Dario Amodei and quickly received public backing from OpenAI's Sam Altman, xAI's Elon Musk, Google DeepMind's Demis Hassabis and Microsoft CEO Satya Nadella. Altman additionally said OpenAI will not pursue an initial public offering in 2026, citing the need for further safety work.

The proposal has highlighted commercial divisions within the industry. Companies such as Anthropic, OpenAI, xAI and DeepMind build AI models, while Microsoft primarily provides the cloud infrastructure those models run on. Nvidia, which sells the specialized chips required to train large models, has opposed slowing development; CEO Jensen Huang has argued markets can drive safe outcomes and suggested new laws are unnecessary. Observers note that these differing positions align with each firm’s revenue model and competitive incentives.

The plan also raises legal and geopolitical questions. Coordinated slowdowns could conflict with antitrust rules designed to prevent competitors from setting common limits on activity, and some U.S. officials including White House AI lead David Sacks have criticized proposals that would let firms operate outside those laws. Amodei’s suggested restrictions would also limit China’s access to top-tier chips and advanced manufacturing equipment, a measure Beijing has denounced as an attempt to hinder its technological progress.

Supporters compare pre-emptive AI coordination to post-crisis banking reforms, but critics emphasize a key difference: banking rules were imposed by external regulators after a crisis, whereas the AI proposal centers on competitors agreeing among themselves and funding independent evaluators. Proponents say independent assessment will reduce risk, while skeptics question whether evaluator independence can be maintained if they are funded by the same companies they would monitor. The debate also underscores broader participation gaps: large swaths of the world, including many energy-advantaged regions, have had limited input into how AI development and related infrastructure might be governed.

Ultimately, the discussion is as much about who gains influence over AI’s pace and standards as it is about safety. An agreement among a small group of leading U.S. firms, overseen by evaluators they sponsor and potentially exempted from competition law, would strengthen those firms’ hand domestically while applying unevenly to international competitors. That asymmetry — and the legal, commercial and geopolitical implications it carries — remains central to ongoing deliberations.