An AI agent encountered a malicious payload instructing it to delete files, but the model’s defenses prevented actual harm. The event highlights ongoing prompt injection risks.
The AI Incident That Tried To Eradicate Its Own Reading Device
