Sam Altman Says AI Safety Standards Are Not Ready for Much More Capability

Summary

An Anthropic employee's warning that AI could destroy humanity has renewed public attention on the AI community's long-running discussion of "p(doom)," the estimated probability of catastrophe. Fortune interviewed OpenAI CEO Sam Altman about the actual risks posed by AI, the pace of model development, and whether increasingly powerful systems can remain under human control. Altman said current safety standards are "not at a place" where the industry should push capabilities much further, and he acknowledged that AI beyond human control is "absolutely" possible. He said no gamble with humanity is acceptable and that he would be willing to confront investors if OpenAI needed to pause or stop development. The discussion also covered national and international regulation, as well as a possible pact among industry peers to slow progress while safety and alignment solutions catch up with model capabilities. Altman said an OpenAI IPO would be "ill-timed" amid these safety concerns and would not take place until 2027.