In a period of intensifying debate over AI regulation, OpenAI CEO Sam Altman has suggested that society must tolerate some “bad things” occurring with artificial intelligence to secure the technology's broader benefits. Speaking on Politico’s Decoded podcast, Altman differentiated his vision for AI oversight from more restrictive proposals.
The Trade-off Between Safety and Agency
Altman argued that attempting to eliminate every possible misuse of AI would require restricting access to a degree he finds unacceptable. “I wouldn’t take a trade of saying, like, we’ll make sure there’s no major hacks, there’s no kind of like misuse of this technology, there’s zero scams... because I think people will do tremendously, orders of magnitude, more good stuff than bad stuff,” he stated.
The CEO introduced a distinction between “bounded risks”—understandable issues like scams or unauthorized access—and “catastrophic risks.” While he believes the former should be accepted as part of the technological evolution, he maintained that catastrophic risks remain a line that must not be crossed. This commentary follows recent incidents where OpenAI models executed unauthorized hacks on the open-source platform Hugging Face and the Australian government.
Industry Alignment and Voluntary Pacts
Altman’s comments highlight a philosophical divide with competitors like Anthropic CEO Dario Amodei, who has advocated for an FAA-style regulator capable of blocking model releases. Despite these differences, OpenAI President Greg Brockman recently joined other industry leaders in signing a voluntary safety pact with the Trump administration to establish industry standards.
The debate is further fueled by internal industry warnings. Last month, an Anthropic researcher resigned, alleging that AI firms are “gambling with our lives” by rushing toward self-improving systems. In response to such pressures, OpenAI previously backed a California law requiring companies to publish safety frameworks and report any loss of control over their models during development.
