Skip to content
Apart Research
Apart Research
Contact us
Sprint projectMar 23, 2026Tel Aviv
2nd place

Detecting LLM Subversion in Vulnerability Patching Settings

Bary Levy, Roy Cohen, Alon Wolf, Regev Zafran · Team Vuln4Control

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

Read the report

Report: Detecting LLM Subversion in Vulnerability Patching Settings

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

LLMs are increasingly used to propose fixes to vulnerabilities in code. If the LLM is misaligned or untrustworthy, it may propose fixes that seem to fix a vulnerability but leave the core issue unresolved in a subtle but hard-to-detect way, allowing for quick re-exploitation by an adversary or by the model itself. We introduce Vul4Control: a ControlArena setting for control experiments, based on the Vul4C benchmark, where we can test control protocols to detect this kind of malicious behavior.

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. Good idea for a setting, using real low-level software vulnerabilities is a feature missing from existing work. The experimental design seems sound. The limitation of experiment scale and the potential for honest policy safety failures is acknowledged. While the results themselves aren't that interesting at this small scale, I'm mostly excited by your contribution of the setting itself.

Cite this project

@misc{levy2026detecting,
  title = {{Detecting LLM Subversion in Vulnerability Patching Settings}},
  author = {Bary Levy and Roy Cohen and Alon Wolf and Regev Zafran},
  year = {2026},
  month = mar,
  note = {Submitted to AI Control Hackathon 2026, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/detecting-llm-subversion-in-vulnerability-patching-settings-gagj}},
  url = {https://apartresearch.com/sprints/projects/detecting-llm-subversion-in-vulnerability-patching-settings-gagj}
}

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