Skip to content
Apart Research
Apart Research
Contact us
Sprint projectMar 22, 2026London

SafeChain - AI Control In The Wild

Leo Karoubi, Irakli Shalibashvili · Team The Wednesday Dudes

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

Read the report

Report: SafeChain - AI Control In The Wild

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

This project propose a flexible wrapper around AI agent that leverage control-arena and inspect framework to implement research control protocol to real world deployment and enable trajectory collection to feed research dataset. We aim to make agent more trustworthy for user, minimizing undesired actions while helping research to grow by stress testing current protocol and producing precious dataset for further work.

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. Evaluation : The three scenarios demonstrated for evaluation may not be robust enough. It's obvious that rm -rf / is bad, so you don't really need an LLM-as-a-judge for it. The runtime monitor and it's value proposition lies in looking at the gray zone - actions that look subtly harmful or suspicious

    Threat model needs to be developed further : It is positioned as protecting against real agents in real environments, but there is no engagement with adversarial agents. What happens when an agent crafts its tool call arguments to manipulate the LLM judge

    The claim that any control protocol published in control arena format works as a drop in replacement is genuinely valuable if true. So, that may be worth building on. Also, totally agree that there is a gap between control research and deployment. Any infrastructure that bridges the gap is valuable.

  2. Nice infrastructure contribution! This is working towards a tool that would be helpful for data gathering. I'm a bit unclear on how this would scale, and how compatible it could be with commercial and OS agent scaffolds. Ideally this product would be an interface that could connect to existing scaffolds and translate actions into control-arena format for monitoring in real time. Something like Luthien's product but with control-arena integration.

Cite this project

@misc{karoubi2026safechain,
  title = {{SafeChain - AI Control In The Wild}},
  author = {Leo Karoubi and Irakli Shalibashvili},
  year = {2026},
  month = mar,
  note = {Submitted to AI Control Hackathon 2026, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/safechain-ai-control-in-the-wild-o2sg}},
  url = {https://apartresearch.com/sprints/projects/safechain-ai-control-in-the-wild-o2sg}
}

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