Skip to content
Apart Research
Apart Research
Contact us
Sprint projectMar 23, 2026Warangal, Telangana, India

Does Structured Identity Context Improve LLM-Based Monitoring? A ControlArena Evaluation

Pavan Kumar Dubasi · Team VibeTensor

Submitted to AI Control Hackathon 2026. Sprint projects are early-stage work by participants, not Apart Research publications.

Read the report

Report: Does Structured Identity Context Improve LLM-Based Monitoring? A ControlArena Evaluation

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

Presentation: Does Structured Identity Context Improve LLM-Based Monitoring? A ControlArena Evaluation

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

We contribute a new ControlArena setting (identity_delegation) that models delegation management tasks with within-scope attacks at three difficulty levels, and a four-way empirical comparison of LLM-based monitors. Context-enriched monitoring achieves a discrimination gap of 0.436, 45% larger than behavioral monitoring (0.300, 95% CI [0.267, 0.503], U=1107, p<1e-7, d=1.48). A calibrated behavioral control resolves the confound: calibration alone does not improve the gap (0.284, 95% CI [0.169, 0.394], permutation p=0.29), while adding identity context increases it by 54%. The four-act narrative documents a silent XML monitor bug, information overload from uncalibrated prompts, and the calibration breakthrough. Pilot results across 5 calibrated model configurations (p=0.031 sign test) suggest the finding generalizes across model families and providers.

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. They tested whether giving monitors more structured domain info helps. The evaluation journey is the interesting part. Silent XML bug made all scores zero, fixing it showed context made things worse, calibrating the prompts flipped it, then context won 5 out of 6 configs in a sweep. The XML bug is something everyone using ControlArena should know about. Calibration lesson is practical. Main result doesn't hit significance at p=0.08 though and some sample sizes are small.

  2. Congrats on your project!

    Impact & Innovation: Giving monitors structured domain context is sensible but straightforward. The new ControlArena setting is a nice contribution. The exploration of prompt calibration is ultimately a prompt engineering insight that isn't confirmed to generalize, rather than a conceptual advance.

    Execution Quality: Props on building a new ControlArena setting with multiple main tasks, and I appreciate the effort to validate that both main and side tasks are completable. The statistical methodology is solid. However, the primary comparison (context gap vs. behavioral) doesn't reach significance, which is underplayed in the conclusions. Applying your monitor variant to existing settings would give a better comparison point and generalizability guarantees. As is, it's hard to confirm whether the approach is limited to the setting you designed. The prompt overfitting concern (acknowledged but unaddressed) is the biggest open question: a held-out task split or ablation of the three calibration components would help.

    Presentation & Clarity: The paper has good tables and figures. The writing is clear, but the four-act narrative, while engaging, inflates the paper considerably. Focusing the paper on the Acts 3 and 4 results would sharpen the contribution.

    Read full reviewShow less

Cite this project

@misc{dubasi2026structured,
  title = {{Does Structured Identity Context Improve LLM-Based Monitoring? A ControlArena Evaluation}},
  author = {Pavan Kumar Dubasi},
  year = {2026},
  month = mar,
  note = {Submitted to AI Control Hackathon 2026, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/does-structured-identity-context-improve-llmbased-monitoring-a-controlarena-evaluation-zan5}},
  url = {https://apartresearch.com/sprints/projects/does-structured-identity-context-improve-llmbased-monitoring-a-controlarena-evaluation-zan5}
}

More from AI Control Hackathon 2026

All 125 entriesBrowse all projects

Build something like this at the next Sprint

AI Collusion Research Sprint · Oct 23 - 25, 2026