Monoculture False Alarms: Collusion Detectors Mistake Shared AI Models for Conspiracy
Maryam Hampaei · Team Singularity Momentum
Submitted to AI Collusion Research Sprint. Sprint projects are early-stage work by participants, not Apart Research publications.
Collusion detectors assume that innocent firms act independently, but AI pricing agents running the same base model react alike without coordinating. We tested 11 standard collusion detectors, including correlation measures, an information-theoretic test (pairwise net influence) and an LLM judge, in a pre-registered repeated pricing game with innocent and seeded colluding pairs. Most detectors falsely flagged about one in three innocent same-model pairs, and almost no innocent cross-model pairs. Net influence, applied to LLM agents for the first time, was protected only when every firm used the same model. Real colluders who agreed a price and held it looked less suspicious than innocent same-model pairs. Setting thresholds separately for same-model and cross-model pairs cut false alarms to 13–21%. Auditors therefore need to know which model each AI agent runs.
Reviews
No public critique yet.
Cite this project
@misc{hampaei2026monoculture,
title = {{Monoculture False Alarms: Collusion Detectors Mistake Shared AI Models for Conspiracy}},
author = {Maryam Hampaei},
year = {2026},
month = oct,
note = {Submitted to AI Collusion Research Sprint, an Apart Research Sprint},
howpublished = {\url{https://apartresearch.com/sprints/projects/monoculture-false-alarms-collusion-detectors-mistake-shared-ai-models-for-conspiracy-cnd5}},
url = {https://apartresearch.com/sprints/projects/monoculture-false-alarms-collusion-detectors-mistake-shared-ai-models-for-conspiracy-cnd5}
}More from AI Collusion Research Sprint
- 1st placeView project: Monoculture Looks Like Collusion: A Replay Test for Hidden Information Flow Between LLM Agents
Monoculture Looks Like Collusion: A Replay Test for Hidden Information Flow Between LLM Agents
Collusion audits ask whether the actions of different agents depend on each other. Agents built on one model act alike without communicating, so these audits accuse honest fleets. Among four honest Claude Haiku agents …
- 2nd placeView project: Looking for a Hidden Agenda: Can a Model’s Activations Reveal Steganography that its Text Hides?
Looking for a Hidden Agenda: Can a Model’s Activations Reveal Steganography that its Text Hides?
Stego-probe
AI agents can collude without saying anything suspicious, just by choosing which true facts to mention. We built a hiring task where Qwen3-8B invents its own covert signals that get past an LLM overseer, then asked …
- 3rd placeView project: Do collusion probes generalise? Prompt confounds in NARCBench
Do collusion probes generalise? Prompt confounds in NARCBench
I investigated HELIOS, an activation-probe for detecting collusion. I found some apparent flaws, and analysed the (acknowledged) unreliable generalisation of the approach.