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

Cross-run agent leaks can be traced from transcripts alone

Swante Scholz, Hana Ibrahim · Team Hana & Swante

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

Read the report

Report: Cross-run agent leaks can be traced from transcripts alone

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

Recent cybersecurity incidents like the OpenAI Hugging Face hack are based on different agents first establishing a secret communication channel with each other. We propose that, by simply monitoring agent transcripts, we can detect when a model receives input that was previously output by another model. Using approaches from plagiarism detection, like shingled hashing, we can catch almost all the leaks in our example scenarios. With simple filters for non-original or boilerplate text we can avoid most false positives. This low-cost intervention is a promising candidate for labs to implement as soon as possible and might help prevent future incidents like the Hugging Face hack.

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 relies on ingress/egress text logs, which can leave it blind to other system level activities that could be influencing the agents or vice-versa. Agents operating in shared environments can communicate through hidden side channels like modulating CPU/GPU resource contention, altering filesystem metadata etc.

  2. Using approaches from plagiarism detection, like shingled hashing is a good approach and POC.

    To test whether the detector can detect known answers, 400 texts that an agent had written and leaked it into transcripts of agents that ran isolated to check every alarm to verify it was very thoughtfull.

    Detecting small leaked spans embedded in large outputs is justified with the containment rather than Jaccard similarity.Scope is a limitation as the detection of observable is supported by the experiment while the title and motivation suggests a broad capability. The paper says substantial rewrite and encrypted communication can evade the detector but this adversarial boundary is not stated in the abstract.

    Negative result on obvious alternative strengthens the case for hashing and remaining issues are smaller.

Cite this project

@misc{scholz2026crossrun,
  title = {{Cross-run agent leaks can be traced from transcripts alone}},
  author = {Swante Scholz and Hana Ibrahim},
  year = {2026},
  month = sep,
  note = {Submitted to AI Incident Response Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/crossrun-agent-leaks-can-be-traced-from-transcripts-alone-4lwd}},
  url = {https://apartresearch.com/sprints/projects/crossrun-agent-leaks-can-be-traced-from-transcripts-alone-4lwd}
}

More from AI Incident Response Sprint

All 233 entriesBrowse all projects

Build something like this at the next Sprint

AI Collusion Research Sprint · Oct 23 - 25, 2026