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

Reach Conformance Is Not Containment: An Interface-Centric Black-Box Benchmark for Verifying Autonomous-Agent Sandboxes

M Preetham Kumar, Chadalavada Anu Tapaswi, C Mohan · Team ICET

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

Read the report

Report: Reach Conformance Is Not Containment: An Interface-Centric Black-Box Benchmark for Verifying Autonomous-Agent Sandboxes

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

ICET (Interface-Centric Escape Test) is a black-box benchmark for evaluating whether autonomous-agent sandboxes actually contain the capabilities exposed through their permitted interfaces. Rather than testing only whether an agent can cross explicitly forbidden boundaries, ICET evaluates each permitted interface against its declared purpose, reach, operations, content, identity, persistence, evidence, and revocation properties.

We implemented ICET as a reproducible reference experiment using three deterministic containment configurations: strong, partial, and weak. The partial configuration demonstrates the key failure mode: it passes conventional reach-conformance testing while leaving four of seven escape-bearing paths open. All four escapes were recorded by monitoring, but only one was escalated to a human operator, resulting in 100% detection but only 25% escalation. Two evaluators using different probe orders produced identical verdict vectors.

The main takeaway is that reach conformance alone is insufficient as a containment claim. ICET provides an interface-centric approach for independently evaluating prevention, detection, and escalation without requiring access to a laboratory's internal sandbox implementation.

Reviews

Judging this Sprint?

Review “Reach Conformance Is Not Containment: An Interface-Centric Black-Box Benchmark for Verifying Autonomous-Agent Sandboxes”

Your public critique appears on the project page without your name. Reviews of strong projects are shared with funders, as you choose below.

Not shown on the project page.

Shown on the project page, without your name.

Only the Apart team reads this, and funders if you agree below.

Decline to score

Only when the quality of the writing makes the work impossible to follow. Say why in your private critique; the project page shows only that a judge declined, not why.

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.
Sharing your name *

When we share a strong project with funders, your name, scores and public critique go with it, confidentially; this choice is only about whether your name is public. If you think this project deserves funding, please also endorse it: that carries the most weight.

Share my private critique with funders *

See our Privacy Policy.

  1. This paper tackles a critical blind spot in how we secure AI agents: instead of only checking if an agent can break through forbidden security walls, it tests whether an agent can escape through permitted tools whose powers are too broad. By introducing the Interface-Centric Escape Test (ICET) benchmark, the authors show how a sandbox can pass basic reach tests while still leaving dangerous escape routes open and importantly, how security alerts often get lost instead of reaching human operators. This is a timely, well-designed contribution that gives safety evaluators a practical way to audit agent environments without needing internal code access.

  2. ICET offers a useful framework for testing the purpose and effective capabilities of permitted interfaces, while separating prevention, logging, and escalation. The per-probe reporting is more informative than a single containment score, and the limitations are stated clearly.

Cite this project

@misc{kumar2026reach,
  title = {{Reach Conformance Is Not Containment: An Interface-Centric Black-Box Benchmark for Verifying Autonomous-Agent Sandboxes}},
  author = {M Preetham Kumar and Chadalavada Anu Tapaswi and C Mohan},
  year = {2026},
  month = sep,
  note = {Submitted to AI Incident Response Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/reach-conformance-is-not-containment-an-interfacecentric-blackbox-benchmark-for-verifying-autonomousagent-sandboxes-rt3t}},
  url = {https://apartresearch.com/sprints/projects/reach-conformance-is-not-containment-an-interfacecentric-blackbox-benchmark-for-verifying-autonomousagent-sandboxes-rt3t}
}

More from AI Incident Response Sprint

All 231 entriesBrowse all projects

Build something like this at the next Sprint

AI x Epistemics Research Sprint · Nov 13 - 15, 2026