Skip to content
Apart Research
Apart Research
Contact us
Sprint projectJan 11, 2026Chicago

Playing Dumb: Detecting Sandbagging in Frontier LLMs via Consistency Checks

James Sullivan · Team James

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

Read the report

Report: Playing Dumb: Detecting Sandbagging in Frontier LLMs via Consistency Checks

Open in a new tab (opens in new tab)
Recording (opens in new tab)Code (opens in new tab)More on huggingface.co (opens in new tab)
Share

I created a dataset where each biosecurity question has a formal "evaluation-looking" version and a casual "normal user" version to detect sandbagging in LLMs. When testing frontier models prompted to sandbag, accuracy dropped over 50% on formal versus casual questions. This shows that consistency checks between question framings can reliably detect when models are hiding their true capabilities.

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. This project tackles an important safety concern: whether models can intentionally underperform during evaluations. The idea of comparing performance across different question framings is intuitive and makes the results easy to interpret.

    The evaluation appears to be conducted on a relatively limited set of questions within a single domain, which makes it harder to judge how broadly the approach will generalize. In addition, the sandbagging behavior is induced through prompting, so it’s not always clear how much this reflects natural model behavior rather than prompt-driven effects.

    It would be useful to test this method across more domains and with additional variations in wording to better separate true sandbagging from general prompt sensitivity. Overall, this feels like a solid starting point for detecting evaluation-aware behavior, with clear potential to be built on further.

Cite this project

@misc{sullivan2026playing,
  title = {{Playing Dumb: Detecting Sandbagging in Frontier LLMs via Consistency Checks}},
  author = {James Sullivan},
  year = {2026},
  month = jan,
  note = {Submitted to AI Manipulation Hackathon, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/playing-dumb-detecting-sandbagging-in-frontier-llms-via-consistency-checks-sntk}},
  url = {https://apartresearch.com/sprints/projects/playing-dumb-detecting-sandbagging-in-frontier-llms-via-consistency-checks-sntk}
}

More from AI Manipulation Hackathon

All 70 entriesBrowse all projects

Build something like this at the next Sprint

AI Collusion Research Sprint · Oct 23 - 25, 2026