OpenAI Boss Sam Altman: The World Must Accept 'Some Bad Things Happening' for AI to Succeed

Sam Altman's comments highlight a fundamental disagreement within the AI industry regarding the balance between innovation, accessibility, and risk. While Anthropic has been a proponent of tighter AI regulation, Altman argues that the potential "orders of magnitude more, good stuff" from AI justifies tolerating certain negative outcomes. He rejects a scenario where all misuse is eliminated at the expense of public access, viewing such a trade-off as "completely unacceptable."

Despite this stance on accepting some harm, Altman does draw a clear boundary for "really catastrophic risks," specifically mentioning "a serious loss of control to AI." This indicates that OpenAI's tolerance for "bad things" is not unlimited and acknowledges a critical threshold beyond which the risks become unacceptable, even for a proponent of wider AI deployment.

The article notes that OpenAI's policy positions have recently begun to align more closely with Anthropic's, despite Altman's verbal distinctions. OpenAI has supported stricter safety requirements, external safety evaluators, and calls to slow the development of advanced AI models. This shift suggests that real-world incidents, such as hacking and unexpected model behaviors, may be influencing the company's approach to regulation, moving it closer to its perceived rival's perspective on safety and control.