Sam Altman has drawn a stark line between OpenAI and Anthropic: not over whether AI carries danger, but over how much danger society should be prepared to live with.

In a Politico Decoded podcast released Monday, the OpenAI chief argued that the technology’s upside will be vast enough that “the world should accept some bad things happening” on the way there. He named hacks, scams and other abuses as costs that will accompany systems capable of enabling people to do “tremendously orders of magnitude more good stuff.”¹

That is an accelerationist bargain, and Altman presented it as a fundamental disagreement with Anthropic, the rival founded by former OpenAI employees after disputes over AI safety. In another account of the interview, he said society should simply deal with “bad things happening” for the technology’s benefits.²

Altman did not argue for abandoning rules altogether. He cast OpenAI as a pragmatic middle ground between Anthropic’s more cautious posture and advocates of near-total deregulation. His warning was that overly restrictive safety policy could concentrate AI in the hands of “a small number of companies or people — or perhaps one country” and become “another way this can go horribly wrong.”¹

But that framing sharpens the clash rather than resolves it. Altman accepts that humans could ultimately lose control of AI systems, yet argues that slowing deployment also carries systemic risks. The opposing safety logic is that scams, hacking and loss of control are not incidental growing pains; they are precisely why powerful AI should face stronger constraints before its capabilities spread further.

The dispute is therefore no longer merely technical. It is a fight over who gets to decide which harms are tolerable—and whether the promise of AI is enough to justify absorbing them first.