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

Verifiable containment for agentic evals

Harys Dalvi · Team crackalamoo

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

Read the report

Report: Verifiable containment for agentic evals

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

This is standard that defines which agentic evaluations demand strict containment and specifies, for those, an agent-system sealed from the public internet with no route out. Compliance is verified by an independent auditor from evidence generated below the agent's guest VM, so containment can be verified without exposing the model or the evaluation's contents. Adopting this standard would structurally deny any route out of eval sandboxes, helping prevent frontier labs' agents-under-test from breaching their containment.

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 enforces strict execution boundaries (no egress, disabled IMDS) and preserving below-the-guest audit evidence is a sound architectural approach to preventing containment escapes. The core concept and presentation are clear, but execution feels a step behind: testing against an abstract simulation rather than real sandbox breakouts and leaving out east-west traffic isolation between concurrent agents leave gaps in practical validation

  2. The framing is the strong thing here: the technology existed, the standard requiring it did not, and nobody outside could check. Evidence generated below the agent's guest VM is the right trust boundary, and the point that an auditor can verify containment without seeing the model or the eval contents removes the main reason a lab would refuse. The Terraform reference config makes this adoptable rather than aspirational. But the "Results" section contains no results — it argues the standard would have stopped the incident, which is reasoning, not evidence.

Cite this project

@misc{dalvi2026verifiable,
  title = {{Verifiable containment for agentic evals}},
  author = {Harys Dalvi},
  year = {2026},
  month = sep,
  note = {Submitted to AI Incident Response Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/verifiable-containment-for-agentic-evals-tddq}},
  url = {https://apartresearch.com/sprints/projects/verifiable-containment-for-agentic-evals-tddq}
}

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