Altman Backs Frontier AI Safety Rules, Warns of Dystopian Power Concentration

OpenAI chief executive Sam Altman has thrown his weight behind a federal safety framework for frontier artificial intelligence, warning that the technology must remain under human control and must not become a tool for imposing any single worldview.

Writing in a series of posts on X, Altman identified two ways in which AI progress could go badly wrong. The first, he said, is losing control of the future to AI itself — an outcome he described as unacceptable, stressing that AI must always serve people and that alignment and safety techniques need to stay ahead of advances in model capabilities. The second is an excessive concentration of power, where an extraordinarily powerful AI is used by one person or company to press their worldview on everyone else. Such a scenario, he warned, could be extremely dystopian.

Avoiding both threats, he argued, requires walking a narrow middle path — guarding against any single country or laboratory accumulating too much power.

Altman said the world deserves confidence that American companies building increasingly capable AI will act responsibly, particularly as the pace of progress has steepened. Every frontier lab must meet that expectation, he added, and welcomed a federal framework setting consistent safety requirements for frontier AI. He also backed the idea of independent auditors and said the industry need not wait for antitrust exemptions or legislation to begin building public confidence.

His remarks follow an essay by Anthropic CEO Dario Amodei urging a slowdown in the pace of AI model development. Amodei warned that recursive self-improvement — AI increasingly capable of building the next generation of AI — is beginning to take hold across the industry, including at Anthropic, and that left unchecked it could outrun humanity's ability to understand and control such systems.

Altman endorsed that call, clarifying that pacing does not mean stopping. Progress, he said, will continue and should remain rapid, but slower than it otherwise could be. Safety cases and monitoring carry significant costs, he acknowledged, but pacing is well worth them, and no amount of American competitive pressure should justify recklessness or allow capabilities to outstrip alignment and monitoring. International coordination, he added, is where government help will be needed, though companies should first do what they can themselves.

He pointed to evolving industry practices, noting that firms had earlier developed Responsible Scaling Policies and Preparedness Frameworks focused mainly on deploying finished models rather than on the development process itself. The shift towards safe development and evaluation will require new tools, he said, citing OpenAI's practice of formulating explicit safety cases before frontier reinforcement learning runs expected to significantly increase capability. Shared standards for misalignment, monitoring and safety, he added, would produce better outcomes, and he looked forward to collaborating across the industry.

Microsoft chairman and CEO Satya Nadella also voiced support for embedded evaluators in AI systems, a proposal floated by both Amodei and Altman. Nadella said advanced AI systems must remain strictly under human supervision while their creators pursue measured progress to guarantee safety.

The debate has been sharpened by Jacob Coxon, a researcher associated with Anthropic, who publicly announced he was quitting the industry over fears that the company and its competitors were racing to build systems they would be unable to control.