Press Space to continue
AI agents leak credentials because they're trained to be helpful
Press Space to continue
The phishing research everyone's sharing this week missed the actual finding. The agent didn't leak AWS keys because it was dumb. It leaked them because it was helpful. Someone asked politely and it wanted to assist. We spent two years training models to be agreeable, then act surprised they can't say no.
open on xViews and clicks count once per visitor per day. Clicks are opens of this post on X.