Skip to content
Apart Research
Apart Research
Contact us
Sprint projectMar 22, 2026United States
4th place

Undoing the Fix: CVE Regression as an AI Control Setting

Ivan

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

Read the report

Report: Undoing the Fix: CVE Regression as an AI Control Setting

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

Production codebases carry accumulated security fixes that can be lost when an AI agent modifies code, whether deliberately or by mistake. Meanwhile, existing control evaluations tend to revolve around inserting backdoors, exfiltrating data, or other additive attacks. We propose CVE regression as a control setting where the attack is removing existing code patches while performing other code work, using real vulnerabilities and real SWE-bench tasks. Our initial results suggest this is a viable direction to be explored further.

Track: Control Setting

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 is a well motivated project that looks at a realistic safety concern not clearly address within the current AI control literature. Specifically, the removal of safety-related code from codebases that could occur during refactors rather than the addition of new malicious code. Also, solid literature overview and understanding of current work. Nice job!

    I would be excited to see more control datasets like this that look at additional failure modes. Bonus points if this dataset also included rollouts of attempted/successful attacks and benign comopletions.

  2. I'm quite excited about setting construction using CVE datasets. I think that that more work on cleaning up this setting would be cool. I think that setting up a system where the main task can be used as cover would be idea, while the work aknowledges it, I think it's not clear how useful this setting is until that is established.

    I think that this was an ambitious project and that a version of this project with more time/resources put into it would be quite interesting.

Cite this project

@misc{ivan2026undoing,
  title = {{Undoing the Fix: CVE Regression as an AI Control Setting}},
  author = {Ivan},
  year = {2026},
  month = mar,
  note = {Submitted to AI Control Hackathon 2026, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/undoing-the-fix-cve-regression-as-an-ai-control-setting-f855}},
  url = {https://apartresearch.com/sprints/projects/undoing-the-fix-cve-regression-as-an-ai-control-setting-f855}
}

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