The sentence
“We believe that the world should accept some bad things happening for the benefits of this technology and people having the agency,” Sam Altman told Politico in an interview published on Monday, reported by The Next Web. He said he would not trade those benefits for a promise of no hacks, scams or misuse, because in his view people will do far more good with AI than harm.
That is a position about ordinary misuse, not about everything. Altman said the world should not accept the worst risks, and named a serious loss of control to AI as the example.
Where he parts from Anthropic
The interview also set out the disagreement between the two labs that has been implicit in a year of competing regulatory proposals. Altman described the view that AI is so dangerous that a single lab should hold it and distribute the gains. He said he understood the argument, and did not agree with it, calling that approach “a completely unacceptable trade-off”.
Anthropic’s answer, given to Politico by a spokesperson, was that its proposed rules apply only to frontier models — the most advanced systems — and not to the wider field. That is the narrower claim, and it is the one Anthropic has made consistently in its public policy work.
The Next Web also reported that Altman supported Dario Amodei’s recent call to slow the development of AI models, which makes the dispute less about pace than about who holds the controls and what the rules reach.
Why the framing matters now
Both companies are talking past a US Congress that has not passed a federal AI statute, while states have. OpenAI has consistently argued for a lighter federal framework; Anthropic has argued for stronger safeguards on the most advanced models. Those are not symmetrical positions, and an interview is where the asymmetry gets stated plainly rather than filed as a comment letter.
The gap also shows up in what each company has already agreed to. OpenAI signed the EU Code of Practice on transparency and this week began watermarking ChatGPT text inside the European Union, a measure it is explicitly not applying globally. Anthropic has pushed for disclosure requirements that attach to a model’s capability rather than to the territory it is sold in. Both are forms of compliance; they imply very different ideas about where the obligation sits.
What to watch is whether the “frontier models only” line holds as a drafting principle. Every proposed federal bill so far has had to define where the frontier sits, and that definition is what decides whether a rule applies to two companies or to two hundred.
