OpenAI CEO Sam Altman told POLITICO’s Decoded that “we believe that the world should accept some bad things happening for the benefits of this technology and people having the agency.”
He drew a boundary around that position. Altman separately said he does not accept “the really catastrophic risks,” including “a serious loss of control to AI,” according to Business Insider’s interview report, published October 4, 2026, and co-written by interviewer Brendan Bordelon.
Asked where OpenAI diverges from Anthropic and its CEO, Dario Amodei, Altman described a disagreement over access, harm prevention and who should control powerful AI. He argued that preserving public agency justifies accepting some harmful uses instead of concentrating the technology in a single institution to prevent them all.
His comments come amid scrutiny of OpenAI-related hacking incidents and rogue model behavior. They also require careful interpretation: Anthropic’s documented regulatory proposals do not demand zero harm, and the reporting describes growing overlap between the companies’ policy positions.
Altman Puts Broad Access Ahead of Preventing Every Harm
Altman’s objection centers on the price of an absolute safety guarantee.
He described a perspective in which AI becomes so powerful and dangerous that “a single lab in San Francisco should have it and make sure nothing bad happens,” while deciding how to distribute its benefits. He called that “a completely unacceptable tradeoff.”
In his framing, restricting access could prevent harmful activity, but it would also give a small group control over what everyone else can do with AI. His concern is about institutional power as well as technology.
Altman connected that argument to major hacks, scams and other harmful activity. He said he would not accept the proposed bargain of eliminating those outcomes at the cost of broad access, because he expects people to do substantially more good than bad with the technology.
That expectation is Altman’s judgment, not a finding established by the interview. The published account provides no measurement of benefits against harms, no method for comparing them and no threshold at which the balance would become unacceptable.
“Agency” is central to his argument. It suggests that letting people decide how to use AI has value beyond any individual application. Under that view, access should not depend entirely on a developer’s confidence that every possible use will be beneficial.
Broad access and harm prevention are not necessarily mutually exclusive, though. The interview’s contrast between public agency and a single controlling lab leaves open which intermediate safeguards OpenAI would support. Deployment restrictions, outside evaluations and incident reporting can be debated without assuming either unrestricted access or institutional monopoly.
