Sam Altman says AI benefits justify accepting some harmful outcomes

OpenAI CEO Sam Altman's argument is that getting the benefits of artificial intelligence means accepting some harm along the way. In an interview with Decoded, he said "the world should accept some bad things happening," according to Business Insider.

That is not an endorsement of every risk. The same report said Altman rejected "the really catastrophic risks" of AI. His position separates harms he believes society can tolerate from outcomes he considers unacceptable.

On October 3, Altman raised a different safety concern: treating AI as a religious force. Axios reported that he was "very uncomfortable" with attributing religious power to models, describing his comments as a veiled shot at Anthropic's work on the soul of AI.

Coverage on October 4 turned to his argument about accepting risks. Business Insider reported his comments to Decoded and his rejection of catastrophic outcomes. Digital Trends also reported that he opposed excessive regulation.

By October 5, the regulatory disagreement was more explicit. According to Forbes, Altman said OpenAI, unlike Anthropic, favours a "lighter-touch regulatory stance" because AI will be used for more "good stuff than bad stuff."

The harms he had in mind included hacks and scams, The Verge reported on October 5. Separately, CNN Brasil reported his argument that AI should remain broadly accessible to the public.

His comments also addressed the industry's political spending. On October 5, The Next Web reported that Altman said money AI companies were pouring into US politics had failed to improve how people viewed the technology.

Altman is making a case for broad access and lighter regulation, not announcing a new regulatory settlement. The central distinction in his remarks is between accepting some harmful uses and accepting catastrophic risks, according to Business Insider and Forbes.

The supplied coverage does not define a precise threshold between those categories. Hacks and scams are concrete examples, but the reports do not set out how much harm would be acceptable or what safeguards would enforce that boundary.

The unresolved issue is how OpenAI's preferred lighter regulatory approach would distinguish tolerable harms from unacceptable risks. The coverage describes Altman's position, but does not provide a detailed proposal or a timetable for putting that distinction into practice.