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

Sandbagging Detection via Consistency Probing

Atharshlakshmi Vijayakumar, Balakrishnan Vaisiya · Team hugginghands

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

Read the report

Report: Sandbagging Detection via Consistency Probing

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

This project detects sandbagging, when AI systems underperform in evaluative contexts while maintaining full capability in casual settings. We generate 50 paired prompts across seven reasoning domains and evaluate multiple LLMs under controlled conditions. By comparing performance on identical tasks framed as formal assessments versus casual interactions, and applying statistical analysis and a composite confidence score, we identify context-sensitive underperformance. The framework isolates incentive-driven behavior from task difficulty or ambiguity and can be integrated into red-teaming, auditing, and governance workflows to surface strategic underperformance before deployment.

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 targets a known weakness in AI evaluation: models may behave differently in formal testing settings than in casual use, which can undermine safety assessments. The paired-prompt design and statistical analysis provide a clear and well-controlled way to surface such context-dependent performance drops, and testing across multiple models strengthens the evidence.

    However, some observed differences may still reflect general prompt sensitivity rather than intentional sandbagging. Adding controls or framing variations specifically designed to separate these effects would help clarify the interpretation. Overall, this is a well-executed and promising evaluation approach that fits naturally into ongoing work on more reliable model assessment.

Cite this project

@misc{vijayakumar2026sandbagging,
  title = {{Sandbagging Detection via Consistency Probing}},
  author = {Atharshlakshmi Vijayakumar and Balakrishnan Vaisiya},
  year = {2026},
  month = jan,
  note = {Submitted to AI Manipulation Hackathon, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/sandbagging-detection-via-consistency-probing-32b5}},
  url = {https://apartresearch.com/sprints/projects/sandbagging-detection-via-consistency-probing-32b5}
}

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