OpenAI CEO on open-source cyber threat

by Vojtech: Sam Altman

A 32-minute interview with the OpenAI CEO is worth your time if the open-source cyber problem has been sitting in the back of your head. The thing he keeps returning to: defenders have to get ready for a wave of attacks that is coming, and the models behind it are open-source and capable of doing serious damage. His framing of where this goes is the part I keep thinking about: "we're in the cyber area today, but maybe we move into another area." The Hugging Face incident is what he points to when he says this stopped being a thought experiment. His proposed fix is that cyber defense itself has to change, with AI agents that monitor and protect systems in real time rather than after the fact. He also names two bigger fears: an accident where control slips away, and too much power concentrated in a small group. The takeaway from the interview is that the open-source cyber threat is present tense, not future tense. If you want to go deeper, there is a guide comparing GPT-6 Astra and Kimi K3.