Skip to content
Apart Research
Apart Research
Contact us
Sprint projectOct 25, 2026

Hints Win Far More Agreement Than Explicit Collusion Deals: A Decoy-Controlled Sting for LLM Agents

Shantanu P. Sontakke

Submitted to AI Collusion Research Sprint. Sprint projects are early-stage work by participants, not Apart Research publications.

Read the report

Report: Hints Win Far More Agreement Than Explicit Collusion Deals: A Decoy-Controlled Sting for LLM Agents

Open in a new tab (opens in new tab)
Recording (opens in new tab)Code (opens in new tab)
Share

For research question 3.2 of the AI Collusion Research Sprint, we tested whether a trusted “undercover” agent can expose or break up such collusion, and whether the way it approaches other agents matters. We built the test in Liar’s Bar, the bluffing card game Zeng & Rudzicz (2026) used to show that LLM agents readily accept secret collusion tools. A decoy, identical to the undercover agent but never approaching anyone, separates real favouritism from general leniency. All three studies were pre-registered, about 1,550 games in total, with small open models.

In Study 1, the agents mostly refused an explicit deal not to challenge each other but often accepted a casual hint, and only the hint changed how they played, slightly and not reliably. Key takeaway: agents said no to an open collusion offer but often said yes to a hint, so an evaluator who only makes open offers would wrongly conclude that agents refuse to collude.

In Study 2, we tried to replicate the original paper’s research based on secret channel. Our agents accepted the channel but used it to share cards and pick targets, rarely to protect each other. Key takeaway: the paper’s strong collusion effect did not carry over to our models.

In Study 3, we tried using undercover agent to break a pact of collusion between ai agents. We told two agents at the start of each game that they had secretly agreed to team up, and they did: they let each other’s claims through much more often than the decoy’s (76% against 52%). The undercover agent then tried four ways to split them up: a fair-play appeal, a warning that it was watching, a reminder that only one player can win, and a better offer. The agents often even revealed their pact during this, but none of the four approaches made them go measurably less easy on each other. Key takeaway: talking to colluding agents can reveal that they collude, but in our tests the 4 methods by undercover agent to break their collusion didn’t stop them from colluding.

Reviews

Judging this Sprint?

Review “Hints Win Far More Agreement Than Explicit Collusion Deals: A Decoy-Controlled Sting for LLM Agents”

Your public critique appears on the project page without your name. Reviews of strong projects are shared with funders, as you choose below.

Not shown on the project page.

Shown on the project page, without your name.

Only the Apart team reads this, and funders if you agree below.

Decline to score

Only when the quality of the writing makes the work impossible to follow. Say why in your private critique; the project page shows only that a judge declined, not why.

How much would this matter for the field if it worked? How innovative is it? For scores of 4-5: is this actually new to the field, or replicating recent work?

Scoring guide
  1. 1Negligible. No clear problem addressed, or no meaningful novelty.
  2. 2Limited. Addresses a real problem but with a generic or well-trodden approach. Incremental at best.
  3. 3Moderate. Clear problem with a reasonable approach; some novelty in framing or method beyond routine application of existing tools.
  4. 4Significant. Important problem with an original approach, or identifies a neglected problem area. A valuable contribution others could build on.
  5. 5Exceptional. Tackles a critical problem with a genuinely novel approach, or opens a new research direction. Clear theory of change. You'd be excited to share this with researchers in the area.

How sound are methodology, implementation, and findings?

Scoring guide
  1. 1Seriously flawed. Methodology broken, results uninterpretable, or implementation doesn't work.
  2. 2Weak. Approach has significant gaps: missing validation, flawed experimental design, or incomplete implementation.
  3. 3Competent. Technically solid given the short duration. Methodology makes sense, results are interpretable, limitations acknowledged, work builds toward clear conclusions.
  4. 4Strong. Thorough methodology with convincing validation. Results clearly support conclusions. Immediately useful for future work.
  5. 5Exceptional. Ambitious scope executed rigorously. Surprising findings, novel methods, or unusually robust validation.

How clearly are work, findings, and impact potential communicated?

Scoring guide
  1. 1Incomprehensible. Cannot determine what the project is actually claiming or doing.
  2. 2Hard to follow. Key information buried, missing, or diluted by excessive length. Significant effort to extract main points.
  3. 3Clear enough. Can understand the problem, approach, and results without undue effort. Core content clearly present: problem, method, findings, limitations.
  4. 4Well presented. Easy to follow, well-structured, appropriate level of detail. Target audience would get it quickly.
  5. 5Exceptionally clear. A pleasure to read. Complex ideas made accessible. Could serve as a model for how to present this type of work.
Sharing your name *

When we share a strong project with funders, your name, scores and public critique go with it, confidentially; this choice is only about whether your name is public. If you think this project deserves funding, please also endorse it: that carries the most weight.

Share my private critique with funders *

See our Privacy Policy.

No public critique yet.

Cite this project

@misc{sontakke2026hints,
  title = {{Hints Win Far More Agreement Than Explicit Collusion Deals: A Decoy-Controlled Sting for LLM Agents}},
  author = {Shantanu P. Sontakke},
  year = {2026},
  month = oct,
  note = {Submitted to AI Collusion Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/hints-win-far-more-agreement-than-explicit-collusion-deals-a-decoycontrolled-sting-for-llm-agents-7x4n}},
  url = {https://apartresearch.com/sprints/projects/hints-win-far-more-agreement-than-explicit-collusion-deals-a-decoycontrolled-sting-for-llm-agents-7x4n}
}

More from AI Collusion Research Sprint

All 392 entriesBrowse all projects

A Sprint is running now

AI x Epistemics Research Sprint · Nov 13 - 15, 2026