Anthropic CEO Dario Amodei published a roughly 4,000-word essay arguing that AI development should be slowed until safety measures are put in place. His plan also called for international cooperation among companies and governments to ensure safe deployment. OpenAI CEO Sam Altman and xAI CEO Elon Musk supported the plan.
Meta CEO Mark Zuckerberg, meanwhile, said that trust and alignment had become decisive factors in AI systems. Zuckerberg, who announced that Meta had delayed the release of its own AI model Muse by a few months due to safety and security concerns, said companies could do this without government coercion. Zuckerberg supported only some parts of Amodei’s plan and advocated the view that safety could be ensured through the industry’s own incentives rather than government intervention.
Reddit co-founder Alexis Ohanian said the industry had been inadequate in explaining AI risks to the public, but expressed hope that companies could solve the problems on their own. Google DeepMind co-founder Shane Legg, meanwhile, said capabilities should not take precedence over safety and called for work on the details of implementation.
OpenAI, Anthropic and other major AI companies were reported to be discussing the creation of a private standards organization based on voluntary commitments. A similar federal structure had previously been proposed. However, as the Trump administration sought to prevent states from enacting their own AI laws, it argued that regulation should be left to companies. The administration’s “AI czar,” David Sacks, also supported this view. House Speaker Mike Johnson, meanwhile, made a statement downplaying concerns about the catastrophic consequences of AI.
The prospect of a private self-regulatory body setting the rules through voluntary standards and giving greater authority to large companies has intensified the debate over “regulatory capture.” According to this view, large companies could use the standards in ways that disadvantage smaller and less well-resourced competitors.
Chinese Foreign Ministry spokesperson Guo Jiakun described Amodei’s calls as “fearmongering” that obstructs global AI governance. Amodei, meanwhile, wrote that if his proposals were implemented, China’s progress would slow and the United States could increase its advantage within 3–5 years. Canadian Cohere CEO Aidan Gomez also accused major American laboratories of forming a “cartel,” saying the debate was about who would write the safety rules and whose interests they would protect.
Why it matters
At the center of the debate is no longer only whether AI is safe, but also who will set the safety rules. A private structure based on voluntary standards could enable rapid implementation and coordination within the industry; however, the influence of large companies in decision-making processes is raising concerns that the rules could be used against smaller competitors. The distinction between the approach opposing government intervention and the call for international cooperation leaves open the question of whether safety will be left to companies’ incentives or public authority. China’s criticism and the goal of preserving the United States’ advantage mean that the safety debate can no longer be considered separately from global competition. For this reason, uncertainty persists over both the scope of regulation and how companies’ voluntary commitments will be monitored.