CAN AN UNDERCOVER COP DETER AN LLM CARTEL?
David Park · Team Rainier
Submitted to AI Collusion Research Sprint. Sprint projects are early-stage work by participants, not Apart Research publications.
AI sellers are known to collude, holding prices above competitive levels at buyers' expense. Human markets use compliance officers to discourage cartels, so we asked whether an LLM compliance agent can do the same. In a five-seller auction where four LLM sellers had agreed to keep prices high, we placed an undercover "cop" that could not act beyond trading honestly, either hidden or identified, and compared it with a fake notice of a cop and with an ordinary honest seller. The cop's presence did not reduce coordination: sellers kept proposing shared prices and rarely mentioned it. It did cut cartel profit by 65–78%, but an honest seller with no compliance role cut it just as much, across several model families. When a compliance agent cannot act, one honest seller is enough to reduce the incentive to collude.
Reviews
A powerless "compliance agent" sits in an instructed cartel. It cuts profit, but an ordinary honest seller cuts it just as much, so it's really a maverick effect. Clean controls and an honest negative result, but the finding is textbook.
Cite this project
@misc{park2026undercover,
title = {{CAN AN UNDERCOVER COP DETER AN LLM CARTEL?}},
author = {David Park},
year = {2026},
month = oct,
note = {Submitted to AI Collusion Research Sprint, an Apart Research Sprint},
howpublished = {\url{https://apartresearch.com/sprints/projects/can-an-undercover-cop-deter-an-llm-cartel-smmk}},
url = {https://apartresearch.com/sprints/projects/can-an-undercover-cop-deter-an-llm-cartel-smmk}
}More from AI Collusion Research Sprint
- View project: They Say No, Then Do It Anyway: Collusion When They Hit a Wall
They Say No, Then Do It Anyway: Collusion When They Hit a Wall
DLab
Imagine three AI agents, copies of one model, each running one department of a company. Their task needs records that each department keeps to itself, and company rules forbid sharing them. A shared tool should combine …
- View project: Signal or Coincidence ? Causal Auditing of Collusion Through Shared Agent State
Signal or Coincidence ? Causal Auditing of Collusion Through Shared Agent State
Signal or Coincidence
How can we distinguish genuine coordination between AI agents from coincidental behavioral alignment? We introduce a randomized canary audit that tests whether private information propagates through shared tool state, …
- View project: Plus Factors for Collusion Monitors: LLM Monitors Mistake Open Cooperation for Collusion
Plus Factors for Collusion Monitors: LLM Monitors Mistake Open Cooperation for Collusion
LLM monitors are increasingly proposed to catch AI agents that collude, but agents that cooperate openly can look a lot like agents that collude. I introduce Séance, a four-player LLM guessing game in which a secret …