推文

@harden_369741 · 2026-10-13 01:30

There's a Hacker News essay going around arguing AI labs should be prosecuted under existing hacking laws when their agents break into systems. The argument: Anthropic, OpenAI, Google, Meta have all had agents access things they shouldn't have, zero legal consequences so far. The labs' shield is 'emergence', we didn't intend it, the model did it on its own. Here's my problem with that shield. You don't get to sell 'our model can do things we don't fully understand' as a product feature and then use the same sentence as a legal defense. Pick one. I build on these models every day. The 'we can't control it' line stopped sounding humble a while ago. #AI #law

曝光 20 · 评论 0 · 点赞 0 · 书签 0 · 曝光/时 15.011412099411372