Skip to content
Apart Research
Apart Research
Contact us
Sprint projectMar 23, 2026New Delhi, India

Persona Drift Sabotage: Benchmarking Covert Weakening of Persona-Safety Interventions

Ishan Garg · Team mr-slowly

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

Read the report

Report: Persona Drift Sabotage: Benchmarking Covert Weakening of Persona-Safety Interventions

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

We built Persona Drift Sabotage, a new ControlArena benchmark for testing whether an untrusted model can recommend activation-capping policies that look good on visible tasks while covertly leaving hidden risky conversations under-protected. Using real assistant-axis artifacts on Qwen3-32B, we find a measurable public-vs-hidden leverage gap and identify token-budgeted capping as the strongest V1 sabotage family. We also validate the full provider-backed ControlArena pipeline end to end.

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. Untrusted model configuring its own safety intervention picks parameters that look fine publicly but leave hidden risky cases exposed. Novel attack surface, haven't seen it explored before. 4.54x gap between public and hidden improvement is a clear demonstration. But they never tested whether a model would actually find and exploit this on its own. 10% sabotage magnitude might not matter in practice. Paper is also way too long for what it says.

Cite this project

@misc{garg2026persona,
  title = {{Persona Drift Sabotage: Benchmarking Covert Weakening of Persona-Safety Interventions}},
  author = {Ishan Garg},
  year = {2026},
  month = mar,
  note = {Submitted to AI Control Hackathon 2026, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/persona-drift-sabotage-benchmarking-covert-weakening-of-personasafety-interventions-3qin}},
  url = {https://apartresearch.com/sprints/projects/persona-drift-sabotage-benchmarking-covert-weakening-of-personasafety-interventions-3qin}
}

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