Google Researchers Reveal Every Way Hackers Can Trap, Hijack AI Agents

A Google DeepMind paper maps six attack categories against autonomous AI agents—from invisible HTML commands to multi-agent flash crashes.