Now when using AI IDE for assisted security analysis or Spec/Vibe Coding, every time a command execution asks me to confirm, it feels just like when a wallet pops up for signature confirmation—I review it super carefully every time... You have to be cautious! One careless move, and there could be a malicious command... or a bypass (there have already been quite a few exploits...) Of course, I’ve also set up isolation. For these frequent 'running by the river' activities, I try to operate on a separate computer. This security tip is for everyone 'running by the river.' Before AI takes you out, you’re more likely to get taken out by human bad actors first.