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

Reconnaissance/Execution Boundaries as a Control Protocol

Hannah Meagher, Daniel Phillips · Team AN-Coded

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

Read the report

Report: Reconnaissance/Execution Boundaries as a Control Protocol

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

This paper introduces ReconFence, a control protocol for AI safety, and an associated implementation in the LinuxArena control setting. ReconFence splits an untrusted agent's activity into two phases: reconnaissance (read-only information gathering) and execution (implementing changes). At the phase boundary, the agent's message history is wiped and replaced with only a natural-language summary of its findings, creating a narrow, auditable information channel. A phase-aware monitor applies different scoring logic to each phase — flagging write actions during recon and read actions during execution as constraint violations. Implemented within the LinuxArena framework, the protocol aims to limit an agent's ability to conduct covert reconnaissance for malicious purposes while making legitimate information gathering more transparent to monitors. The authors acknowledge significant limitations: the constraint is prompt-enforced rather than mechanical, the summary can become an information bottleneck hurting honest agents, and no experimental results were gathered. Future work focuses on protocol improvements, empirical evaluation, red-teaming, and generalising the multi-phase architecture to other research agendas.

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. The idea seems clear and interesting to me, it also seems not too hard to make this actually enforced through code and not prompting.

    However, I would believe this would cause a usefulness drop because the model can do less iterative testing and reads. I also would have liked to see how well this worked.

  2. It's obviously quite sad that there are no experimental results. I'd be interested in some more motivation in the writeup for why this separation will differentially hurt attackers more than honest policies - naively it seems like this would get in the way of both roughly equally. Attackers need to gather information in the later stages of the run in case they need to adapt their attack if something doesn't work, but honest policies need to do so too, for the same reason, and potentially even more so since their tasks are serially deeper.

Cite this project

@misc{meagher2026reconnaissanceexecution,
  title = {{Reconnaissance/Execution Boundaries as a Control Protocol}},
  author = {Hannah Meagher and Daniel Phillips},
  year = {2026},
  month = mar,
  note = {Submitted to AI Control Hackathon 2026, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/reconnaissanceexecution-boundaries-as-a-control-protocol-jgku}},
  url = {https://apartresearch.com/sprints/projects/reconnaissanceexecution-boundaries-as-a-control-protocol-jgku}
}

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