Sam Altman: world should accept some AI harm

The exchange came in Decoded by POLITICO's newsletter and podcast, with POLITICO senior technology reporter Brendan Bordelon asking the questions. Altman framed his position as a disagreement over worldview rather than a technical dispute, saying a fundamental difference remains with Anthropic. Asked where he and Dario Amodei split, he pointed to a trade-off: considering AI's benefits and people keeping their agency, the world should accept that AI causes some bad things.

He said he understands the perspective of those who want tight control, including the idea that a single company should hold the technology and make sure nothing bad happens. But he said that clashes with OpenAI's advocacy for lighter AI regulation. Altman specifically refused a condition of guaranteeing that large-scale hacking, technology misuse, fraud and other wrongdoing never occur, saying he believes people will do far more good than bad. At the same time, he stated that catastrophic risks, including a severe loss of AI control, are not acceptable.

The test for OpenAI is likely to be where that line sits in concrete rules, since Altman has ruled out both zero-incident guarantees and tolerating catastrophic outcomes. How regulators and rival labs read that middle ground — especially those pushing stricter controls — is likely to shape the next round of AI policy debates.