AI Agents May Have a Memory Problem

A new study by researchers at Princeton University and Sentient shows it's surprisingly easy to trigger malicious behavior from AI agents by implanting fake "memories" into the data they rely on for making decisions.