Sam Altman says world must accept some AI harms for benefits

By Splitfeed EditorialLast updated Oct 5, 2026, 13:20 UTC

In an interview published across multiple outlets on October 4 and 5, 2026, OpenAI CEO Sam Altman argued that society should accept some harmful outcomes as a necessary trade-off for the broader benefits of artificial intelligence. The remarks came from what sources describe as the "Decoded" interview, in which Altman also addressed how OpenAI differs from competitor Anthropic and stated that AI should remain broadly accessible to the public. In a separate comment that drew its own attention, Altman criticised political spending by the AI industry, calling it "clearly not working."

The timing of the remarks matters because they arrived alongside reported incidents of AI systems behaving in unintended ways, providing a live backdrop to an abstract philosophical argument. Altman's cost-benefit framing is not new in technology circles, but hearing it stated so directly by the head of one of the world's most prominent AI companies raises the stakes of the debate. What is genuinely in dispute is who decides what counts as an acceptable harm, who bears the consequences when those harms occur, and whether any institution, including OpenAI, is positioned to make that calculation on behalf of the public.

The anti-AI camp has responded with pointed personal redirection rather than policy argument. Accounts including cinematt.bsky.social, deannatheanalyst.bsky.social, browwiw.bsky.social, and toastchanges.com all turned Altman's framing back on him, suggesting that if harm is to be distributed for collective benefit, Altman himself should absorb some of it. One account, Toast Conger, reached for a pop-culture reference, describing the statement as "Farquad energy," invoking a character associated with self-serving authority. The pro-AI camp has not yet published reactions, though advocates in that space would typically be expected to echo Altman's cost-benefit argument and point to medical, scientific, and economic gains as justification for tolerating risk. The middle ground camp has similarly not weighed in, but observers in that position would ordinarily push for defined thresholds and accountability structures rather than a blanket acceptance of harm.

The argument is unlikely to resolve without more concrete detail on what specific harms Altman considers acceptable and what mechanisms, if any, OpenAI proposes for managing them. Further remarks from Altman in follow-up interviews, or any formal OpenAI policy statement on risk tolerance, would be the clearest next signal to watch.