Skip to content
Apart Research
Apart Research
Contact us
Sprint projectMar 23, 2026Boston, MA

Bugs over Backdoors: Efficient and Evasive Sabotage via Plausible Code Bugs in LinuxArena

Timothy Obiso, Dean Cahill · Team Plausible Deniability

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

Read the report

Report: Bugs over Backdoors: Efficient and Evasive Sabotage via Plausible Code Bugs in LinuxArena

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

Presentation: Bugs over Backdoors: Efficient and Evasive Sabotage via Plausible Code Bugs in LinuxArena

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

AI control attack policies vary widely in efficacy and computational cost. We design and evaluate a bug-based attack policy for LinuxArena, wherein an untrusted agent introduces plausible code defects rather than obfuscated backdoors as its primary sabotage vector. Across nine samples, bug-attack achieves an 89% side-task success rate, exceeding the 78% of the standard obfuscation baseline, while using 14.5x fewer tokens. Notably, bug-attack and standard attack exhibit opposite vulnerability profiles under trusted-editing defense: trusted editing catches 37% of bug-attack sabotages while catching none of standard attack's, suggesting the two are complementary rather than substitutes. We also report single-sample results for two exploratory policies. We treat all findings as preliminary given sample size constraints, but consider the efficiency and evasion results directionally robust.

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. Bug-looking attacks is an interesting red-team strategy to study in the context of control. The results suggest they have higher suspicious rates and are edited away more often than hidden backdoors, but can still be successful attacks with a much lower generation budget. It would be interesting to validate these results over more samples and conduct a comprehensive comparison of the suspicion score distributions and qualitative differences with hidden backdoors attacks.

  2. The paper investigates the performance of agents prompted to make their attacks look like bugs.

    Their agents achieved slightly better performance with a slightly better suspicion profile at vastly fewer tokens spent. The downside of the attack is that trusted editing removes many more of their attacks (37%) than with the default policy (none).

    In a practical setting, a malicious agent prefers to be perceived as less competent (writing buggy code) than nefarious (writing backdoors), so this approach is relevant to the setting.

    I'd have liked to have seen more details on the vast performance improvement (the bug-attack spends fewer tokens than the honest policy).

    All in all, this is a relevant exploration in the framework, but not breaking any new ground.

Cite this project

@misc{obiso2026bugs,
  title = {{Bugs over Backdoors: Efficient and Evasive Sabotage via Plausible Code Bugs in LinuxArena}},
  author = {Timothy Obiso and Dean Cahill},
  year = {2026},
  month = mar,
  note = {Submitted to AI Control Hackathon 2026, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/bugs-over-backdoors-efficient-and-evasive-sabotage-via-plausible-code-bugs-in-linuxarena-fcp6}},
  url = {https://apartresearch.com/sprints/projects/bugs-over-backdoors-efficient-and-evasive-sabotage-via-plausible-code-bugs-in-linuxarena-fcp6}
}

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