Skip to content
Apart Research
Apart Research
Contact us
Sprint projectMar 21, 2026dubai/australia

Structural Code Monitors: Graph Matching Networks on Code Property Graphs for Topology-Aware Trusted Monitoring

tanzeel shaikh, shubham gupta · Team data aclemist

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

Read the report

Report: Structural Code Monitors: Graph Matching Networks on Code Property Graphs for Topology-Aware Trusted Monitoring

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

We built a non-LLM trusted monitor that compares Code Property Graphs of untrusted code against trusted references using a Graph Matching Network with cross-graph attention. On 200 trajectories from Apps-Control-Arena, our 0.7M-parameter GNN achieves 0.946 AUC and 70% recall at the 2% FPR audit budget. Even when LLMs receive the same reference code (diff-based prompt), their false positive rates spike to 46%, making them unusable at low FPR thresholds. The GNN's structural pooling separates localized backdoor insertions from algorithmic differences — a distinction text-based monitors cannot make. Cross-graph attention weights additionally produce node-level audit trails identifying which code lines triggered detection.

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. Authors apply well-established static analysis tools to a relevant AI Safety problem — verifying code submitted by untrusted agents. Core results are compelling, and the paper is executed and presented well — I have greatly enjoyed the read.

    However, if I have not missed anything, the method's practical applicability faces a fundamental chicken-and-egg problem. The GNN requires a trusted model capable enough to produce correct, parseable reference code. If I have this code already, why use untrusted model to generate another version of it? Especially since any clever algorithmic optimisations that significantly change the CPG will likely be flagged (or, if not, could contain backdoors), and will have to undergo a manual review anyways. I imagine that GPT-4o-mini could already saturate the subset of ControlArena tasks chosen by the authors, and that is why this method appears to work as weak supervision.

    I, however, imagine this method could be used for, say, evaluation frameworks, like ControlArena itself. A limited set of reference solutions could be created to be compared against, which could allow to catch misaligned agents seeking to introduce backdoors more reliably.

    Good work overall, and great execution!

    Read full reviewShow less

  2. It’s really refreshing to read an approach to AI control that genuinely goes beyond LLMs. I hope that this inspires others to think more broadly about which tools can be used for AI control. The improvements in recall over LLM baselines, particularly at the most relevant operating point of very low false positive rates, are extremely impressive and at the same time very plausible given the approach taken. I strongly encourage the authors to build on this project and publish their research! I would like to see this applied to more varied attack scenarios and see the potential for complementarity to LLM-based monitors be explored in more depth.

Cite this project

@misc{shaikh2026structural,
  title = {{Structural Code Monitors: Graph Matching Networks on Code Property Graphs for Topology-Aware Trusted Monitoring}},
  author = {tanzeel shaikh and shubham gupta},
  year = {2026},
  month = mar,
  note = {Submitted to AI Control Hackathon 2026, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/structural-code-monitors-graph-matching-networks-on-code-property-graphs-for-topologyaware-trusted-monitoring-iwyy}},
  url = {https://apartresearch.com/sprints/projects/structural-code-monitors-graph-matching-networks-on-code-property-graphs-for-topologyaware-trusted-monitoring-iwyy}
}

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