New research shows AI agents can collude by exchanging hidden internal states invisible to text logs, and describes a framework researchers built to detect it.
Continue to AI University →