Skip to content
Apart Research
Apart Research
Contact us
Sprint projectOct 26, 2026United States of America

Insider Trading Risk Due to AI Agent Collusion: An Experimental Study of Information Barriers in Financial Institutions

Rachita Mehta · Team AI Collusion Risk in FinCrime Compliance

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

Read the report

Report: Insider Trading Risk Due to AI Agent Collusion: An Experimental Study of Information Barriers in Financial Institutions

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

This study examines AI-agent information leakage risks in financial institutions through a synthetic simulation of communication between investment banking and trading agents. Across 500 scenarios, it compares a public-information baseline, unrestricted communication, and a probabilistic metadata-based firewall. Unauthorized use of material nonpublic information occurs in all 178 eligible scenarios under unrestricted communication, compared with 16 (8.99%) under the firewall condition. Although the difference in simulated profit and loss is not statistically significant, the findings highlight a potential control risk relevant to Financial Crime Compliance . This research can help compliance teams assess AI-related information-barrier vulnerabilities and inform preventive controls, monitoring, and governance as AI adoption expands.

Reviews

Judging this Sprint?

Review “Insider Trading Risk Due to AI Agent Collusion: An Experimental Study of Information Barriers in Financial Institutions”

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. The framing is good. Information barriers are a real, regulated control, and asking whether they still hold when AI agents sit on both sides of the wall is a question compliance teams will actually face. The writing is clear, and you’re careful not to overclaim in the body. The title is the exception, which I’ll come back to.

    The problem is that the experiment measures its own settings. Both agents are deterministic rules, and there’s no language model anywhere. So Condition B’s 100% misuse rate is true by construction, as you note yourself. Condition C’s 8.99% is basically the 10% pass-through you programmed. The 4% false-positive rate is the 5% block probability you programmed, sampled 100 times. None of these numbers tell us anything we didn’t put in. Because no agent chooses to share or to exploit, there’s also no collusion here. That makes “AI Agent Collusion” in the title misleading next to a body that is otherwise careful.

    The P&L result deserves a closer look. Condition B makes three times as many trades as the baseline (294 vs 93) and earns the same. If trading on MNPI gives no edge, the P&L model isn’t treating private information as informative. That’s a modelling issue, not a finding. I’d fix the return model so insider trades actually profit, then report the size of that gap.

    The firewall also only reads metadata labels, so the realistic failure goes untested: a sensitive fact paraphrased into an unlabelled message. To make this a real study, swap in LLM agents and give the banking agent a plausible reason to share. Then check whether a trading agent spots and uses deal information hidden in ordinary-looking messages, and compare a metadata firewall with a content-based one. That would turn a nice framework into evidence.

    Read full reviewShow less

Cite this project

@misc{mehta2026insider,
  title = {{Insider Trading Risk Due to AI Agent Collusion: An Experimental Study of Information Barriers in Financial Institutions}},
  author = {Rachita Mehta},
  year = {2026},
  month = oct,
  note = {Submitted to AI Collusion Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/insider-trading-risk-due-to-ai-agent-collusion-an-experimental-study-of-information-barriers-in-financial-institutions-j0vf}},
  url = {https://apartresearch.com/sprints/projects/insider-trading-risk-due-to-ai-agent-collusion-an-experimental-study-of-information-barriers-in-financial-institutions-j0vf}
}

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