On October 4, 2026, Politico published an interview with OpenAI CEO Sam Altman that was quickly picked up by Reuters and others. The line driving the discussion: the world "should accept some bad things happening for the benefits of this technology," which reports quote with slightly different endings -- one adds "and people having the agency."
According to Reuters' account, as carried by US News, Altman said he sees significant "daylight" between OpenAI and Anthropic, and rejected the idea that "a single lab in San Francisco" should hold the technology to make sure nothing bad happens, calling that "a completely unacceptable trade-off." He framed OpenAI's position as favoring broad access and user agency over tighter control.
Why it is being debated. The two companies' public policy positions have converged this autumn. Reuters notes that Altman said he was on board with Dario Amodei's September call to slow the push toward the most powerful models, and that OpenAI has joined Anthropic in supporting state laws with tougher safety standards and a bipartisan House measure that would place independent safety evaluators inside leading AI firms. This site covered Amodei's "We Must Pace the Frontier" essay on September 12 and the industry split that followed. So the interview reads to some as a reassertion of difference just as the policy gap closes, and to others as a candid statement of a trade-off everyone in the industry makes but few say out loud.
I think Altman is right that zero harm is not an available option, and that concentrating control in one lab has its own risks. But "accept some bad things" is only a policy once you say which bad things, how many, and who decides. Cars, drugs and aircraft are all allowed to cause some harm; what makes that acceptable is that the harms are measured, reported and bounded by someone other than the manufacturer. That is exactly what the House measure OpenAI now backs would start to provide. So the honest version of Altman's position is not a contrast with Anthropic at all -- it is an argument for the independent evaluators and incident reporting both companies say they support. The test of the interview is whether OpenAI pushes those mechanisms through, or uses "acceptable harm" as a reason to keep the definition in-house.
What to watch: whether Anthropic responds publicly, whether the House evaluator bill moves, and whether OpenAI publishes any threshold for what it considers an acceptable harm.
Sam Altman told Politico on October 4 that society should accept some AI harms for its benefits and that there is daylight between OpenAI and Anthropic on risk, even as OpenAI backs the same state laws and House evaluator bill Anthropic supports -- making the real question who measures and bounds the 'acceptable' harm.