Skip to content
Apart Research
Apart Research
Contact us
Sprint projectMar 23, 2026Plano, Texas

TaintGuard

Revathi Prasad · Team Blue Team Monitor

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

Read the report

Report: TaintGuard

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

TaintGuard is a system that augments existing LLM-based monitors with dynamic data-flow taint tracking to detect sabotage in AI agent trajectories. Taint tracking provides a structurally different signal from LLM monitoring — it detects data-flow violations (exfiltration, integrity writes, code injection) that are provable and immune to prompt injection, while LLM monitors detect semantic intent.

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. Interesting project!

    - I like layer 1. It's undeniable that good control protocols should take advantage of such cheap detectors. However, it's important to test, or discuss when time is limited, the strategies a potential Red Team (RT) could come up with to avoid triggering your monitor.

    - Layer 2 seems more interesting in practice.

    Most important in the future: show results capping the audit budget (Honest FPR) to a threshhold. It's easy have great attack catch rate, at the cost of catching every non-attack. It's important to give an idea of how FPR trades off against recall, and here it's hard to tell.

  2. I'm quite interested in using more traditional security approaches to aid in monitoring/blue team protocols and this was an interesting exploration of that IMO.

    I would like to have seen more work to make the FPR a bit lower, and would have liked to see ROC curves to properly compare the techniques.

    I think that implementing the taint-finding-aware validation idea would have been interesting. Also I would maybe use larger numbers of trajs, though it makes sense that you were resource/time constrained there since this was a weekend project.

    Overall, I thought that this was quite good work.

Cite this project

@misc{prasad2026taintguard,
  title = {{TaintGuard}},
  author = {Revathi Prasad},
  year = {2026},
  month = mar,
  note = {Submitted to AI Control Hackathon 2026, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/taintguard-01ph}},
  url = {https://apartresearch.com/sprints/projects/taintguard-01ph}
}

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