As artificial intelligence grows more powerful, a debate is intensifying over how much control governments should have over its development. OpenAI CEO Sam Altman has repeatedly warned that highly advanced AI could eventually become difficult for humans to control. Recently, some researchers in the industry have left their jobs to garner more attention to this possibility, calling for stronger government oversight of frontier AI.

On Sept. 21, OpenAI proposed that countries work together to evaluate the capabilities and risks of advanced AI models and establish international safety standards. The company specifically identified recursive self-improvement – a process in which an AI system could increase its own capabilities with limited human assistance – as an area requiring careful monitoring and safeguards.

Other major AI companies, including Anthropic and Google DeepMind, have also called for greater international cooperation and more active intervention in recent months. Demis Hassabis, co-founder and CEO of Google DeepMind, proposed in July that the United States lead efforts to establish international standards for frontier AI. More specifically, he suggested creating a new standards body modeled on the Financial Industry Regulatory Authority, or FINRA, which oversees the U.S. securities industry.

Not everyone agrees that stronger government regulation is the answer. David Sacks, a technology adviser to President Donald Trump, said concerns about advanced AI are warranted but that some political figures exaggerate the threat. Speaking to CBS News, Sacks argued that AI development should continue and that technology companies should bear primary responsibility for managing its risks.

Trump has also opposed international control of AI. On Sept. 22, he declared before the United Nations General Assembly that the U.S. rejects attempts to establish international authority over AI.

The debate continued on Sept. 23, when Altman and Anthropic CEO Dario Amodei addressed the U.N. Security Council and called for stronger AI safeguards. At the same time, Altman warned against falling into “doomerism,” a mindset rooted in chronic pessimism and fatalism, and argued that a middle path between unrestricted development and stopping AI progress exists. He called for countries to cooperate on standards that would enable “accurate and speedy” reporting of AI-related incidents and greater sharing of safety information.