Sam Altman says society should accept AI harm like scams and hacks for its benefits

In the same interview - part of Politico's Decoded series - Altman said he wouldn't agree to a deal guaranteeing there would be no major hacks, no abuses of the technology, and no fraud, because he believes people will do "tremendously - orders of magnitude more - good stuff than bad stuff" with AI. That calculus has a hard floor, though: Altman said he would not accept "a serious loss of control to AI" ^[1], and separately called a future where a single lab controls the technology "completely unacceptable" ^[1]- a line that reads as pointed given it's his own lab making the call. The distinction Altman draws is specific: tolerate low-grade, recoverable harm (scams, hacks, everyday misuse) in exchange for broad access and "agency," but treat irreversible loss of control as the one line that can't move. He undercut his own reassurance on that front in the same breath: "We have not solved alignment. I believe no lab has solved alignment," he said, adding that he gets "nervous" when he hears rumors that other labs believe they have solved it well enough to keep training unsupervised ^[2].

Altman's comments land just days after David Robinson - OpenAI's safety transparency lead for three and a half years, who oversaw safety reports on twelve frontier model launches - resigned ^[3]. Robinson didn't leave quietly: shortly after, he published an essay in The Atlantic titled "I Quit OpenAI Because Its Culture Is Broken" ^[3], arguing "I agree with other recently departed staff that the companies building this technology aren't being nearly careful enough" and that he is "joining a parade of former colleagues - at OpenAI and the industry's other leaders - who have decided that the current path is unacceptable" ^[4]. Set against that backdrop, Altman's admission that "no lab has solved alignment" reads less like candor and more like confirmation of exactly what his own former safety staff are saying publicly - the reassurance and the resignation are describing the same gap from opposite sides.

The remarks also arrive amid heightened regulatory scrutiny of how OpenAI and its peers secure their models and AI agents following recent security incidents ^[1]. Politically, the reaction was immediate: Senator Chris Van Hollen responded by arguing that instead of accepting harm as a cost of innovation, lawmakers could instead hold companies accountable for the harm they cause - reframing Altman's trade-off as a deflection from corporate liability rather than a philosophical stance on progress. Public reaction skewed overwhelmingly skeptical: the recurring argument across online discussion was that the benefits of the trade-off concentrate at the top - AI companies and their executives - while the costs, the scams, hacks, and misuse Altman named, are externalized onto everyone else. That asymmetry, more than the specific examples of harm, was the center of gravity for most critics.

This wasn't a one-off. In an earlier interview, weeks before the Politico piece, Altman was responding to Treasury Secretary Scott Bessent's G20 criticism that the AI industry had done a 'terrible job' explaining its benefits to the public - and agreed, saying the industry as a whole had handled this horribly, and that OpenAI itself did a bad job too. In that same conversation, Altman again drew a clear red line around catastrophic risk while leaving room on everything short of it - a two-tier framing close to what he'd later repeat to Politico. Read together, the two interviews suggest a deliberate communications strategy: trade candor about near-term harms for credit on long-term benefits. The scale of the public pushback to the Politico remarks, including direct pushback from a sitting U.S. senator, suggests that strategy is, so far, generating more scrutiny than goodwill.