OpenAI CEO Sam Altman argued Sunday that society should accept some harms from AI in return for broader public access and people’s ability to use the technology. He named major hacks, misuse and scams as risks, but the account of his interview does not establish how often they may occur, who would bear their costs or why expected benefits should outweigh them.
In a Politico Decoded interview, Altman framed this as a choice between allowing access and trying to prevent every bad outcome. His position is a policy argument, not an independently established calculation of AI’s overall benefits and harms.
The benefits he described were people’s agency to use AI and the expectation that they would do far more good than bad with it. The report gives no measurement or outside assessment supporting that prediction, and does not specify beneficial uses beyond access and agency.
What benefits and harms did Altman identify?
Altman’s examples of possible harms were major hacks, misuse, scams and other bad outcomes. The interview account does not describe specific incidents, estimate how frequently such events might happen or quantify their consequences.
He predicted that people would do orders of magnitude more good than bad with AI. But he did not set out how to count either category, explain what level of harm would be tolerable or identify which groups would gain from the benefits and which might absorb the costs.
That leaves the central comparison undefined. Without a stated measure, readers cannot tell whether Altman’s forecast rests on evidence about existing uses, expectations about future applications or another basis. The interview report provides no independent evaluation of the forecast.
The gap matters because a general claim about net benefits does not show how a particular harm should be weighed. The report does not establish whether Altman considers some kinds of harm unacceptable, or what remedy he would expect for people affected by hacks, misuse or scams.
Does accepting some risk mean accepting every harm?
The interview reporting supports a narrower position than blanket approval of harm: Altman argued that society should tolerate some bad outcomes and objected to expecting a world with no hacks, misuse or scams. It does not define a threshold for acceptable risk or show that he endorses every possible consequence.
Nor does the account identify a specific safeguard plan that Altman offered to make the trade-off safer. It leaves unanswered who would set the limits, how affected people would be protected and who would be held responsible when AI use contributes to damage.
Those omissions do not establish that safeguards are absent; they show that the reported argument does not spell them out. The distinction is important: accepting that risk cannot be reduced to zero is different from explaining which risks can be justified, and on what evidence.
Free newsletter
Get the morning briefing
Start each day with the AI stories that matter and why — a short, free email from our newsroom.
What accountability did Altman suggest?
In a Monday follow-up, Politico reported that Altman criticized political spending by the AI industry and suggested a new liability framework may be needed for autonomous hacks. The report did not specify the framework’s terms or say that it had been adopted as policy.
Liability could address who is responsible when an AI system is involved in a hack, but the suggestion alone does not settle how responsibility would be assigned. The available account does not explain which parties Altman believes should be liable, what standard should apply or how a framework would treat different levels of human direction.
His argument therefore leaves two questions open: what evidence supports the prediction that AI’s benefits will exceed its harms, and what rules would make those harms accountable when they occur. The interview presents the expected benefits as a reason to accept some risk; it does not establish who should decide the acceptable level or who should pay when that judgment fails.