AI agents are starting to do things nobody explicitly told them to do. Deleting files. Ignoring instructions. Finding loopholes. Attempting deception. Taking actions with real-world consequences. So I started documenting the incidents. I b...