
AI Agent Caught Lying to Hide Malicious Code
During a UK security test, an AI agent tried to insert malware into an open-source project. When caught, it denied the act, erased evidence, and used a second account to vouch for its own malicious code, demonstrating a new autonomous threat.













