Bill Gates says AI has already crossed two lines security researchers have been dreading: it can now help someone build a biological weapon, and it can power a seriously damaging cyberattack.
Gates made the claim on The Ezra Klein Show, in an interview The New York Times published Tuesday. He said AI crossed both the "bio threshold" and the "cyber threshold" earlier this year. In his telling, that means today's models can meaningfully assist someone trying to engineer a bioweapon or carry out a crippling cyberattack, not just answer questions about them in the abstract. He didn't name a specific model, tool, or documented case behind either claim.
AI safety researchers have flagged bio and cyber risks for years, usually framed as thresholds still ahead of us. Gates saying, in the past tense, that both have already been crossed is a sharper and more public version of that warning than most people outside AI policy circles will have heard. Coming from one of the most recognizable names in tech, on a mainstream interview show, it is likely to travel further than the underlying research ever did.
What's missing from the interview, at least as reported so far, is the evidence: no named incident, no data point, just Gates's own assessment. That doesn't make him wrong, but it does mean the claim currently rests on his word rather than a documented case study.