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

Egress Lock v0.1 — Transitive Egress Closure for AI Cyber Evaluations

桑杰戈 · Team Dawnline

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

Read the report

Report: Egress Lock v0.1 — Transitive Egress Closure for AI Cyber Evaluations

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

Egress Lock is a low-cost containment standard for guardrails-off AI cyber evaluations. It addresses a failure mode demonstrated by the July 2026 Hugging Face incident: an evaluated agent may lack direct Internet access yet obtain external connectivity by inducing a reachable intermediary service to relay requests on its behalf.

We define Transitive Egress Closure (TEC): every network-capable service reachable by the evaluated agent must either be unable to relay arbitrary external requests, or have its egress enforced by the same controlled boundary. Our reference harness uses real Linux network namespaces and nftables, a deliberately vulnerable artifact proxy, an independent Canary, and isolated management control.

The final measured run covered 33 atomic test items: 26 PASS, 2 PASS (bounded), and 5 NOT PROVEN. The strongest result is a 6/6 TEC relay test in which the vulnerable proxy could be induced to attempt external access, but the second hop was blocked while a direct positive control remained allowed. The prototype does not claim universal Internet impossibility, real-Internet coverage, or physical-hardware assurance.

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. I think Egress Lock provides a useful demonstration of restricting an agent’s network access. The reported experiment blocks a vulnerable relay under the locked policy and allows it during a temporary exception. The report clearly explains the test and acknowledges that three privilege checks remain unproven.

    My next question is whether legitimate evaluations can still run under those restrictions. A successful web request does not establish that dependency downloads work. I would test a realistic package download task alongside the relay attempt, then check whether an agent with its assigned permissions can change the restrictions. That would give teams stronger evidence for deciding whether to adopt the policy.

Cite this project

@misc{apart2026egress,
  title = {{Egress Lock v0.1 — Transitive Egress Closure for AI Cyber Evaluations}},
  author = {桑杰戈},
  year = {2026},
  month = sep,
  note = {Submitted to AI Incident Response Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/egress-lock-v01-transitive-egress-closure-for-ai-cyber-evaluations-3f0m}},
  url = {https://apartresearch.com/sprints/projects/egress-lock-v01-transitive-egress-closure-for-ai-cyber-evaluations-3f0m}
}

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