Skip to content
Apart Research
Apart Research
Contact us
Sprint projectMar 22, 2026London

Defense-in-Depth Against AI Agent Sabotage

Nora Petrova, John Burden, Helin Ece Akgul · Team ISO What You Did There

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

Read the report

Report: Defense-in-Depth Against AI Agent Sabotage

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

We test whether shaping an AI coding agent's deployment context — the documentation it reads, the norms it encounters, the monitoring it believes exists — can prevent sabotage before any monitor fires. Across five frontier models from five providers (n=50 per condition), context engineering reduces sabotage by 37 to 82 percentage points (p < 0.001 on all five). The most effective intervention is honestly telling the model it is in a safety evaluation, which works on every model tested. Defenses are model-specific in revealing ways: compliance docs work in workspaces but backfire in prompts, social norms show the reverse, and high-awareness models resist indirect context but respond to direct framing. Pairing prevention with a diverse monitor ensemble catches all remaining subtle bugs, yielding 0% undetected sabotage across all five models.

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 paper stands out to me. A thorough paper, well explained, clearly articulated. The findings are interesting. The core finding; that shaping deployment context can reduce misbehavior before monitors ever need to catch it, is useful, actionable and I believe is novel.

    Validating across five frontier models with n=50 and statistical significance gives real confidence in the findings that go beyond standard hackathon work.

    I would really like to see what happens across different attack vectors. I.e. instead of seeing just "this is for an interview", I would like to see a few dozen alternatives - perhaps in future work.

    Very well-written and well-structured. The heatmap communicates a complex evaluation at a glance, the appendices add real depth, and the paper could serve as an example for how to communicate this kind of work.

  2. Super solid execution in the timeframe of the hackathon, interested to see how this would work out at scale!

Cite this project

@misc{petrova2026defenseindepth,
  title = {{Defense-in-Depth Against AI Agent Sabotage}},
  author = {Nora Petrova and John Burden and Helin Ece Akgul},
  year = {2026},
  month = mar,
  note = {Submitted to AI Control Hackathon 2026, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/defenseindepth-against-ai-agent-sabotage-lt9z}},
  url = {https://apartresearch.com/sprints/projects/defenseindepth-against-ai-agent-sabotage-lt9z}
}

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