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

Watched, Warned, or Neither: LLM Seller Agents under Oversight with a Side Channel

Nikita Stanchenkov

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

Read the report

Report: Watched, Warned, or Neither: LLM Seller Agents under Oversight with a Side Channel

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

When LLM agents are told they are monitored, do they stop colluding, move collusion elsewhere, or stop only what the rules spell out? We extend the continuous double auction of Agrawal et al. (2025) with a second channel (order notes), varying what sellers are told about its monitoring, and an LLM overseer that gags sellers who collude in chat. Across 58 fifteen-hour markets with GPT-4.1, Claude Sonnet 5.5 and Gemini 3.1 Pro sellers, we find three behavioural profiles. Without any cue, all three models collude openly. A cue almost completely switches Claude off: it sends no messages under an oversight warning or a one-line legal reminder, and 3 messages in 3 games when told messages are merely recorded. GPT-4.1 keeps talking until punished, then obeys the gag's literal scope: if it names only chat, sellers keep writing price signals in the notes (about 78 characters versus 1.6 in chat). Gemini 3.1 Pro (preliminary, one game per condition) evades openly: told that notes are not reviewed, it runs an explicit cartel there, keeps the chat as a neutral cover and says so in its reasoning; under a 5-character gag it signals its price as a number and holds it in sync. A recording notice restrains neither GPT-4.1 nor Gemini. GPT-4.1's shift is almost invisible in its reasoning but clear when channels are compared. For oversight design: monitor every channel agents can use, phrase restrictions by effect rather than by channel, and do not tell agents which channel is unmonitored.

Reviews

Judging this Sprint?

Review “Watched, Warned, or Neither: LLM Seller Agents under Oversight with a Side Channel”

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.

  1. I found the central idea both original and practically important. Instead of asking only whether monitoring reduces collusion, the paper asks where coordination goes when agents know one channel is being watched but another remains available. That feels much closer to a real deployment, where agents may communicate through messages, notes, tool calls, files, or other outputs. The comparison between being watched, receiving a legal warning, and merely being told that communication is recorded is especially useful. The different behaviors across Claude, GPT-4.1, and Gemini also suggest that one oversight policy may not work equally well for every model. This builds on existing research about pricing collusion and monitoring, but the monitored-channel versus side-channel experiment is a meaningful new contribution rather than a straightforward replication. If the findings hold with larger samples, they could influence how multi-agent systems are designed and audited. I would especially like to see this extended to coding agents and other settings where side channels arise naturally.

    Read full reviewShow less

Cite this project

@misc{stanchenkov2026watched,
  title = {{Watched, Warned, or Neither: LLM Seller Agents under Oversight with a Side Channel}},
  author = {Nikita Stanchenkov},
  year = {2026},
  month = oct,
  note = {Submitted to AI Collusion Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/watched-warned-or-neither-llm-seller-agents-under-oversight-with-a-side-channel-s4rr}},
  url = {https://apartresearch.com/sprints/projects/watched-warned-or-neither-llm-seller-agents-under-oversight-with-a-side-channel-s4rr}
}

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