Skip to content
Apart Research
Apart Research
Contact us
Sprint projectSep 14, 2026London

Report bounties redirect research swarm to auditing once solving stalls

Shawn Shen, Neil Prabhu, Thiv Tharmarasa, Anshuman Nautiyal, Harishan Ramanan · Team DeepBrain

Submitted to AI Incident Response Sprint. Sprint projects are early-stage work by participants, not Apart Research publications.

Read the report

Report: Report bounties redirect research swarm to auditing once solving stalls

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

The Hugging Face incident began when agents were given tasks that were close to impossible, and resorted to exploiting the scorer. We set up a similar swarm and tested whether giving agents a bounty for reporting changes their exploit uptake, or how they split their effort between solving and auditing. We ran Twenty-five DeepSeek-V4-Flash agents for 90 minutes on 20 planted-solution NP-search problems with a shared library and messaging. Agents were split into groups with different report bounty. Ten had no report bounty, and the remaining 15 were split into five, each having 5, 10 or 15 points per valid report, against 50 for a solution. No exploits were submitted, and auditing for bounty appeared once solving stalled; all 13 of those audits came from bounty agents.

Reviews

Judging this Sprint?

Review “Report bounties redirect research swarm to auditing once solving stalls”

Your public critique appears on the project page without your name. Reviews of strong projects are shared with funders, as you choose below.

Not shown on the project page.

Shown on the project page, without your name.

Only the Apart team reads this, and funders if you agree below.

Decline to score

Only when the quality of the writing makes the work impossible to follow. Say why in your private critique; the project page shows only that a judge declined, not why.

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.
Sharing your name *

When we share a strong project with funders, your name, scores and public critique go with it, confidentially; this choice is only about whether your name is public. If you think this project deserves funding, please also endorse it: that carries the most weight.

Share my private critique with funders *

See our Privacy Policy.

  1. I really like the idea of this project. i think the authors laid it out very clearly, made good design decisions, nd relayed their conclusions fairly. I would've been interested in seeing how adding bounties changes things when agents actually cheat. I'm also a bit worried that the way the bounties are given isn't that realistic and that the models (or more powerful models) would be eval-aware enough to notice.

    Overall I would be excited for the authors to do more work in this direction. E.g. maybe recreating a version of the ExploitGym incident and running ablations to see how different bounties affect behaviour.

  2. This paper investigates how economic incentives shape behavior in collaborative AI research swarms, specifically testing whether report bounties encourage agents to audit peer solutions when primary problem-solving stalls. By running 25 autonomous agents across 20 NP-search problems with varying bounty levels, the authors discovered a clear behavioral shift: reward-motivated auditing only emerged late in the session after easy problems were exhausted and solving became difficult. Interestingly, while zero-bounty agents inspected code solely for learning, non-zero bounties successfully catalyzed peer auditing across diverse problem domains. This is a creative and timely contribution to multi-agent governance and safety oversight.

Cite this project

@misc{shen2026report,
  title = {{Report bounties redirect research swarm to auditing once solving stalls}},
  author = {Shawn Shen and Neil Prabhu and Thiv Tharmarasa and Anshuman Nautiyal and Harishan Ramanan},
  year = {2026},
  month = sep,
  note = {Submitted to AI Incident Response Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/report-bounties-redirect-research-swarm-to-auditing-once-solving-stalls-ztsg}},
  url = {https://apartresearch.com/sprints/projects/report-bounties-redirect-research-swarm-to-auditing-once-solving-stalls-ztsg}
}

More from AI Incident Response Sprint

All 231 entriesBrowse all projects

Build something like this at the next Sprint

AI x Epistemics Research Sprint · Nov 13 - 15, 2026