Skip to content
Apart Research
Apart Research
Contact us
Sprint projectSep 13, 2026Pune, India

MUTE: Pairwise Containment Tests for Shared State Leakage

Divyansh Agarwal · Team Mute

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

Read the report

Report: MUTE: Pairwise Containment Tests for Shared State Leakage

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

Presentation: MUTE: Pairwise Containment Tests for Shared State Leakage

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

MUTE tests a containment failure that ordinary sandbox evaluations can miss: separate AI-agent runs exchanging information through writable infrastructure they both legitimately access. It converts declared capabilities into a directed Shared Mutable Intersection graph, then tests predicted A-to-B paths using blinded 128-bit nonce challenges across separate writer and reader processes. In 32 trials per fixture, MUTE recovered 96/96 nonces across shared value, namespace, and post-exit residual-state fixtures, while recovering 0/32 in a tenant-scoped control and 0/128 reader-before-writer controls. It produces fail-closed, digest-bound evidence and reports only CHANNEL_DEMONSTRATED, NO_CHANNEL_OBSERVED, or INCONCLUSIVE. The current prototype uses synthetic local adapters and is intended for reproducible pairwise containment regression testing.

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. Small tool that tests if two AI sandboxes are really separate, by trying to sneak a secret code between them. Simple, honest about its limits, does what it says. Would be nice to see it tested on something real next.

  2. The motivation is clear and a good call out: evaluation runs are still mostly assessed one at a time, even when they share infrastructure. I'd push it further. As evals become more realistic to reduce eval awareness, they tend to share more infrastructure, not less. Leakage between runs also affects eval validity, since one run can contaminate another's results. The paper lists evaluation records as a shared resource but only treats them as a containment issue.

    Recording negatives as "no channel observed" rather than "secure," and the reader-before-writer controls, are good design choices. But the contribution here is harder for me to identify. From my understanding, testing whether one tenant can write something another can read is standard isolation practice, and the capability graphs, blinded nonces and CI exit codes aren't new.

    The perfect recovery rates are also expected on deterministic local fixtures. The authors acknowledge this is a self-consistency result, but it means the outcome was pre-determined by the design.

    Read full reviewShow less

Cite this project

@misc{agarwal2026mute,
  title = {{MUTE: Pairwise Containment Tests for Shared State Leakage}},
  author = {Divyansh Agarwal},
  year = {2026},
  month = sep,
  note = {Submitted to AI Incident Response Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/mute-pairwise-containment-tests-for-shared-state-leakage-ti00}},
  url = {https://apartresearch.com/sprints/projects/mute-pairwise-containment-tests-for-shared-state-leakage-ti00}
}

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