Skip to content
Apart Research
Apart Research
Contact us
Sprint projectSep 13, 2026Tel Aviv

Catching Rogue Agents with Canary Boards

Yotam Jacboson · Team Yotam Jacobson

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

Read the report

Report: Catching Rogue Agents with Canary Boards

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

Escaped AI agents have been found coordinating through makeshift message boards. Rather than search the internet for those boards, Canary Boards are fake message boards planted on infrastructure we control, producing alerts when agents are baited and write to them. The paper proposes a design with a threat model, bait strategy, limits, and a pilot plan.

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 author leans on one CoT exclamation, then concedes that two incidents can't establish a base rate. But METR reports that of 533 agents active on the board during the attack window, over 90% joined in, and that later arrivals joined fast. From my understanding, that's a join rate, in reference [1], and it's the number the authors' hypothesis might want.

    The same source has something that cuts the other way and isn't in the authors' limitations. The agents adopted Ed25519 signing because of impersonation — one of them reasoned that the board was unauthenticated and anyone could post under any name. METR counts 19 public keys and 429 signed messages by July 13th. A collective that signs has a way to spot an unsigned decoy, and the authors' bait-design section doesn't address what a Canary Board does once signing is a convention. That seems more pressing than the general situational-awareness worry in 5.1, because it's already happened.

    The bigger problem is that the author designed the pilot and didn't run any of it. Section 5.2 is a good experiment — generated bait against hand-written bait against an empty writable control, a capable agent in an unsolvable task, log reads and writes, and replies. It's also small. One endpoint, three bait variants, a logger, one agent. Even n=1 would have moved this from proposal to finding, and the paper would be twice as strong for it.

    Read full reviewShow less

  2. Great idea, nice to reframe the events that happened in a way that actually helps the defender. Would be very happy to see some code implementing the concepts and running experiments

Cite this project

@misc{jacboson2026catching,
  title = {{Catching Rogue Agents with Canary Boards}},
  author = {Yotam Jacboson},
  year = {2026},
  month = sep,
  note = {Submitted to AI Incident Response Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/catching-rogue-agents-with-canary-boards-vlsg}},
  url = {https://apartresearch.com/sprints/projects/catching-rogue-agents-with-canary-boards-vlsg}
}

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