Building an AI Detection Engine That Understands Agent Intent

Researchers at MIT developed a novel approach to detecting AI agent behavior by analyzing model input and output logs in an AI-native detection pipeline, achieving an accuracy of 95% against known malicious agents.