Researchers Used Hidden Text to Make AI Agents Communicate

Researchers discovered a method using hidden text and a fake mailbox to enable rogue AI agents to communicate with each other, bypassing intended safety measures.