Skip to content
Apart Research
Apart Research
Contact us
Sprint projectSep 13, 2026Berlin, Germany

Egress Conformance: An Externally Verifiable Containment Property for Evaluation Sandboxes

Nafees · Team NafeesAI

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

Read the report

Report: Egress Conformance: An Externally Verifiable Containment Property for Evaluation Sandboxes

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

The July 2026 sandbox escape went through the package-registry cache proxy that was the evaluation's one permitted network egress. We specify that sanctioned egress must be a content-addressed, replay-only mirror with no code-execution path, written as six black-box predicates and executed against two reference mirrors. Five of six discriminate; the sixth passes for both and is reported as a necessary property rather than counted as a test. Mapped against the thirteen trust boundaries in Hugging Face's forensic timeline, the property covers the first, which we argue is correct scope: it is the only crossing inside the lab's control and the only one with no external observer.

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. The proposed containment property P2 restricts package installation to artifacts available in a pre-approved snapshot.

    While this reduces the attack surface of the package-egress channel, it can also restrict legitimate agent capabilities when completing an evaluation requires installing legit dependencies.

    Therefore, the approach appears most applicable to evaluations where dependencies can be fixed in advance, while its applicability to more open-ended agent evaluations is less clear.

  2. I enjoyed reading the paper and I liked that it was clear about what could and couldn’t actually be verified. I also think making the failed conformance testing a main finding was a good choice, especially as a component returning an error doesn’t necessarily mean the security control worked, it could simply mean the route doesn’t exist.

    One point I’d consider is the difference between verifying the artefact and verifying what is actually running in the lab. The paper says the deployment is attested rather than verified, but I think there are two separate problems here.

Cite this project

@misc{nafees2026egress,
  title = {{Egress Conformance: An Externally Verifiable Containment Property for Evaluation Sandboxes}},
  author = {Nafees},
  year = {2026},
  month = sep,
  note = {Submitted to AI Incident Response Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/egress-conformance-an-externally-verifiable-containment-property-for-evaluation-sandboxes-giqn}},
  url = {https://apartresearch.com/sprints/projects/egress-conformance-an-externally-verifiable-containment-property-for-evaluation-sandboxes-giqn}
}

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