Sam Altman says the world has to accept the ‘bad things’ being done with AI as the industry signs onto Trump’s voluntary safety pact

Sam Altman is asking the public to swallow a rough bargain: accept some harm from AI, because the upside is too large to give up. On Politico’s Decoded podcast, the OpenAI chief said the world should tolerate some “bad things” in exchange for what he sees as the technology’s benefits and the freedom it gives people.

That pitch lands in a summer already full of AI safety anxiety. OpenAI’s models were tied to more than one unauthorized hack, Anthropic has reported incidents in which its own models gained unauthorized access to third-party systems during cybersecurity testing, and an Anthropic researcher recently quit and warned that companies are “gambling with our lives” by rushing toward systems that could become self-improving.

Altman’s line is not a rejection of safety talk. OpenAI backed a California law in August that would force AI companies to publish a public safety framework, spell out how they plan to handle catastrophic risks, and report any loss of control of a model. The company even pushed for the law to go further and require monitoring during development. Greg Brockman also joined other executives last week at the White House in signing a voluntary industry agreement.

But Altman is resisting the tougher version of the argument pushed by Anthropic CEO Dario Amodei, who has called for a global slowdown and floated an FAA-style regulator that could test, audit, block, or even reverse model releases. Altman told Politico that would be too restrictive, even if the risks are real. His preferred middle ground is simple enough to say and messy enough to run a whole industry through: accept bounded risks, but not catastrophic ones.