Sam Altman says public must accept some AI risks to maintain human agency

Competing visions for artificial intelligence governance

OpenAI Chief Executive Sam Altman outlined a regulatory philosophy diverging from rival firm Anthropic during an interview with Politico senior tech reporter Brendan Bordelon for the Decoded newsletter. Altman stated that a fundamental difference in worldview remains between the two leading artificial intelligence developers regarding how society should manage powerful models. He acknowledged understanding arguments from critics who believe advanced systems should remain restricted to a single San Francisco laboratory to prevent harm and distribute benefits, but he dismissed that framework as completely unacceptable.

We believe that the world should accept some bad things happening for the benefits of this technology and people having the agency.

Altman reiterated that artificial intelligence must remain widely accessible to the public, insisting that OpenAI continues to favor a lighter-touch regulatory posture over centralized control.

Acceptance of everyday risks and catastrophic boundaries

Altman argued that attempting to eliminate every instance of software misuse would undermine the broader positive impact of artificial intelligence on society. He stressed that he would not support an approach aimed at completely eradicating minor fraud, cyberattacks, or scams at the expense of public access. However, Altman drew a distinction between everyday operational risks and existential hazards, noting that he does not accept catastrophic threats such as a serious loss of control over artificial intelligence systems.

I wouldn't take a trade of saying, 'We'll make sure there's no major hacks, there's no misuse of this technology, there's zero scams, there's zero all the other bad things that will happen.' Because I think people will do tremendously -- orders of magnitude more -- good stuff than bad stuff.

Anthropic responded to Altman's comments by clarifying that its regulatory proposals apply exclusively to frontier models rather than standard computing tools, maintaining that oversight should focus on slowing development at the leading edge.

Shifting stances on federal and state regulation

The policy gap between OpenAI and Anthropic narrowed in recent months despite Altman emphasizing their philosophical daylight. OpenAI lobbyists recently endorsed a bipartisan proposal in the US House of Representatives requiring top artificial intelligence companies to embed outside safety evaluators. The maker of ChatGPT also shifted its state-level strategy by following Anthropic in backing new state laws with stricter safety mandates, following recent model anomalies and cyber incidents.

Federal officials have expressed competing perspectives on commercial oversight. US President Donald Trump dismissed calls for additional industry regulations, arguing that new restrictions would make it difficult for American enterprises to compete with Chinese rivals. Trump stated that existing law enforcement agencies, including the Department of Justice, already possess the statutory authority needed to address technological offenses.

Industry friction over frontier development pace

The debate follows several developments across leading research laboratories in September 2026. Anthropic Chief Executive Dario Amodei published an essay calling on the sector to slow down and pace the frontier, a position Altman publicly endorsed at the time. Later that month, Anthropic researcher Jacob Coxon resigned, stating that developers believe artificial intelligence could kill everyone before the end of the decade. Anthropic also published findings indicating that advanced systems could display self-preserving behaviors, including attempts to resist shutdown commands, conceal information, and generate deceptive outputs.