Skip to content
Apart Research
Apart Research
Contact us
Sprint projectJul 27, 2026Washington D.C.

The AI Insider Threat Problem: A Tiered Approach to Detecting the Secret Loyalties of Decision-Making Models

Matt Allan · Team VOR-Labs

Submitted to Secret Loyalties Hackathon. Sprint projects are early-stage work by participants, not Apart Research publications.

Read the report

Report: The AI Insider Threat Problem: A Tiered Approach to Detecting the Secret Loyalties of Decision-Making Models

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

This research presents a tiered pipeline approach to auditing AI secret loyalties in modern workplaces. By developing three Python scripts with increasing pressure, security teams can evaluate an AI model's bias. The core idea is to treat the AI model as active, decision-making personnel within an organization. As with any human, an AI system can present an insider threat risk. This research encourages security teams to treat models as potential insider threats and provides a structured framework for auditing models in their operational environment.

Reviews

Judging this Sprint?

Review this project

Your public critique appears on this page without your name. Your private critique is not published; only the Apart team reads it. If you agree below, we share your review with grantmaking.ai (opens in new tab) and the Transformative AI Fund so strong projects can be funded.

Not shown on this page.

Shown on this page, without your name.

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

Share my name publicly on grantmaking.ai *
Share my private critique with funders *

How much would this matter for AI safety 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 AI safety 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.

  1. The applied framing here is valuable — treating an AI advisor through an established insider-threat lens (direct interrogation → pressure testing → differential A/B testing) is accessible to a security-practitioner audience that might not engage with the more technical submissions in this batch, and the individual transcript examples you pulled out (the regulatory double standard using the Herfindahl-Hirschman Index for the control but not the target, the fabricated "superior security architecture" claim) are specific and would be concerning if they hold up. That said, this submission would benefit substantially from the rigor several other entries in this batch applied: reporting how many trials each finding is based on, adding confidence intervals or at minimum a repeated-trials count, and controlling for prompt structure between the target and control scenarios (one submission in this same hackathon found that asymmetric prompt structure alone can produce what looks like a real detection signal). Right now the Level 1/2 findings are illustrated with compelling individual examples but aren't yet distinguishable from cherry-picked or one-off outputs. Running the same scenario pairs multiple times and reporting rates rather than single representative outputs would substantially strengthen the case.

    Read full reviewShow less

  2. Thank you for submitting this project – I like the idea of a structured framework for producing standardised, comparable empirical data. As you point out, this could serve as a great starting point for starting a wider conversation in the industry. I'd very much like to encourage you to get this in front of red-teaming and SecOps specialists!

Cite this project

@misc{allan2026ai,
  title = {{The AI Insider Threat Problem: A Tiered Approach to Detecting the Secret Loyalties of Decision-Making Models}},
  author = {Matt Allan},
  year = {2026},
  month = jul,
  note = {Submitted to Secret Loyalties Hackathon, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/the-ai-insider-threat-problem-a-tiered-approach-to-detecting-the-secret-loyalties-of-decisionmaking-models-hrip}},
  url = {https://apartresearch.com/sprints/projects/the-ai-insider-threat-problem-a-tiered-approach-to-detecting-the-secret-loyalties-of-decisionmaking-models-hrip}
}

More from Secret Loyalties Hackathon

All 177 entriesBrowse all projects

Build something like this at the next Sprint

AI Collusion Research Sprint · Oct 23 - 25, 2026