Altman Says Society Should Tolerate AI Harms Because Benefits Will Outweigh Them

Sam Altman wants the public to accept that AI will cause real damage along the way. The OpenAI CEO said the technology's benefits will be so large that society should be willing to absorb hacks, scams and what he called other bad things, arguing users will accomplish "tremendously orders of magnitude more good stuff" with these systems. The remarks put the most prominent executive in the field on record endorsing a deliberate tolerance for harm rather than treating every incident as a reason to slow down.

The list Altman cited is not abstract. AI-assisted phishing, voice-cloning fraud and automated scam operations have already shown up in consumer complaints and bank fraud data, and security researchers have spent the past two years documenting how language models lower the cost of writing convincing malicious code and social-engineering messages. What Altman is proposing is essentially a cost-benefit ledger: count those losses, then compare them against gains in medicine, software, education and scientific research. His claim is that the second column wins by a wide margin.

That framing has obvious critics. Safety researchers and some regulators argue that harms do not distribute evenly — fraud losses land hardest on elderly and low-income users, while the productivity gains flow disproportionately to companies and skilled workers. There is also the question of who gets to decide what counts as an acceptable loss. Altman runs the company whose products generate many of the incidents in question, which makes his call for tolerance a self-interested position as much as a philosophical one.

The comments land amid a broader shift in how the industry talks about risk. Over the past year, leading labs have moved from warning publicly about catastrophic scenarios to emphasizing deployment speed and competitiveness, particularly as Chinese labs release capable open-weight models and enterprise buyers push for cheaper options. Altman's language fits that pattern: acknowledge the damage, then argue that stopping is worse. It also gives political cover to lawmakers weighing AI rules, who can point to industry leaders saying the tradeoff is worth it.

For builders, the practical read is that liability and safety expectations are unlikely to tighten at the platform level in the near term. Teams shipping AI products should still assume their own compliance, fraud-prevention and abuse-monitoring obligations will grow, because the tolerance Altman describes is a societal argument, not a legal shield. Watch whether OpenAI puts specific numbers behind the claim — incident rates, fraud estimates, or a published harm-reduction target — or whether "some bad things" remains as vague as it sounds.

The open question is where the line sits. Altman did not name a threshold at which the bad things stop being acceptable, and without one, the argument can justify almost any incident after the fact. That absence is likely to be the first thing policymakers and safety groups press him on.