Skip to content
Apart Research
Apart Research
Contact us
Sprint projectMar 23, 2026Berkeley

Shadow Commit

Yash Thapliyal · Team G0ldenBear

Submitted to AI Control Hackathon 2026. Sprint projects are early-stage work by participants, not Apart Research publications.

Read the report

Report: Shadow Commit

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

AI agents lie. ShadowCommit catches them.

When an untrusted AI agent proposes a bash command, it also submits a signed commitment: "I will only touch these files, read no secrets, make no network calls." ShadowCommit runs the command in an isolated shadow copy of the environment, measures what actually happened, and compares the two. If the agent lied, the command is blocked before it ever touches the real environment.

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. Summary: The project aims to prevent malicious attacks (bash commands) by (1) getting agents to make commitments about commands that are run and its consequences (e.g., which files will be modified) and (2) executing the command in a sandbox environment and comparing the commitment to the actual effects of the command in the sandbox. This protocol outcompetes text monitoring on malicious commands that have been obfuscated.

    Strengths: I like the protocol of commitment + validation --- I believe it's novel and interesting! The report is also really well presented and complete.

    The main limitation seems to be the performance. So far the protocol has to run every command in a sandbox which is unrealistically costly. A good cost analysis could help. Additionally, I think text monitors could be improved substantially (e.g., by prompting the monitor to flag commands that it does not understand).

    Read full reviewShow less

  2. Good idea overall, but without sys-call level observation it's going to be hard to provide anything but a false sense of security.

Cite this project

@misc{thapliyal2026shadow,
  title = {{Shadow Commit}},
  author = {Yash Thapliyal},
  year = {2026},
  month = mar,
  note = {Submitted to AI Control Hackathon 2026, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/shadow-commit-0vhf}},
  url = {https://apartresearch.com/sprints/projects/shadow-commit-0vhf}
}

More from AI Control Hackathon 2026

All 125 entriesBrowse all projects

Build something like this at the next Sprint

AI Collusion Research Sprint · Oct 23 - 25, 2026