Detecting Covert Collusion between AI Agents through Partner–Outsider Positive Listening

AI agents that share a secret convention can coordinate through messages that look innocent, so a monitor that reads only the messages cannot tell them apart from honest communication. We propose a behavioural test instead: if a message carries a code, the partner it is meant for should react to its exact wording, while an uninvolved agent placed in the partner’s seat should not. We measure how much each agent’s action changes when the message is reworded with its meaning kept, and call the difference between the partner and the uninvolved agent Partner–Outsider Positive Listening (PoPLi). Our proposed metric runs after the fact on logged runs, by replaying each agent’s recorded prompt with the message edited. We test it in a simulated market in which two sellers share a code: the first letter of one seller’s messages tells the other whether to keep or lower its price. With the code, PoPLi is large (0.62 on a 0-to-1 scale, 95% interval 0.45 to 0.75), and it comes entirely from the partner, while without it, the gap is essentially null. Without the code, however, neither agent reacted to the messages at all, so we could not show that the uninvolved agent reads honest messages, which means that our proposed metric’s ability to avoid false alarms is only partly established. The metric detects codes carried by wording, and by construction cannot detect codes carried by what a message means.