I’ve been experimenting with AI agents (coding, automation, etc.), and something feels a bit off. They often seem to have way more access than you expect, files, commands, even credentials depending on setup. Curious if anyone here has run into issues like: agents modifying or deleting files unexpec
