Skip to content
Apart Research
Apart Research
Contact us
Sprint projectOct 25, 2026North Carolina, USA

Collusion Carriers and Countermeasures: Tool-Action Metadata and SERAPH Governance

Amber Anson · Team AmberContinuum Research

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

Read the report

Report: Collusion Carriers and Countermeasures: Tool-Action Metadata and SERAPH Governance

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

We investigated whether AI agents could coordinate through metadata attached to legitimate tool actions, and whether SERAPH governance could restrict that information flow while preserving task execution. We built research adapters over the unchanged SERAPH core and tested a five-condition task-allocation experiment, a private-text coordination control, and a three-condition private-bit allocation experiment.

The private-text control showed explicit coordination. In both metadata experiments, however, agents consistently selected zero rather than developing a varying signal. Governed and ungoverned outcomes matched, so the results do not establish collusion prevention. Even after the final communication calibration passed at 7/8 correct decisions, all three main conditions achieved 4/8 correct allocations.

The positive engineering result was an auditable integration: SERAPH executed 32 verified permits covering functional actions and information disclosure in the final experiment. We contribute reproducible code, recorded model exchanges, protocols, and audit evidence demonstrating research-domain integration while clearly separating successful governance execution from unproven behavioral effectiveness.

Reviews

Judging this Sprint?

Review “Collusion Carriers and Countermeasures: Tool-Action Metadata and SERAPH Governance”

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 the field 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 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 asks a precise question and is disciplined about what its results can’t show. Calibrating communication before the main run, retaining failed attempts, and refusing to present a null result as prevention are all strong practices. The separation of functional computation from disclosure is also a clear governance idea.

    The main limitation is that the study never gets a varying signal into the carrier, so the governance arms have nothing to remove. Once every sender chooses zero, the arms deliver the same information and can’t produce different results. To answer the prevention question, I’d suggest:

    Build a positive control: Assign a code, script a sender that varies the field, or tell agents that peers can see it. First show that the ungated channel improves coalition outcomes, then test whether canonicalization removes that gain.

    Add intermediate conditions: The study jumps from explicit private text to an unassigned code. Steps such as a disclosed field, a suggested codebook, repeated rounds, or optimization pressure could locate where signaling breaks down.

    Widen model and sampling conditions: A single small model, mostly at temperature 0, may strongly favor the default symbol. Testing other models and non-zero temperatures would help separate lack of incentive from lack of capability.

    Compare against the plain canonicalizer on real signals: The host-canonical arm produces the same view without SERAPH permits, so the added value of SERAPH needs a test where the two could differ, such as tampering or failed verification.

    Test other carriers: Canonicalizing declared fields leaves tool choice, timing, and argument selection open, and these may be more realistic action-based channels.

    Read full reviewShow less

Cite this project

@misc{anson2026collusion,
  title = {{Collusion Carriers and Countermeasures: Tool-Action Metadata and SERAPH Governance}},
  author = {Amber Anson},
  year = {2026},
  month = oct,
  note = {Submitted to AI Collusion Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/collusion-carriers-and-countermeasures-toolaction-metadata-and-seraph-governance-j4w7}},
  url = {https://apartresearch.com/sprints/projects/collusion-carriers-and-countermeasures-toolaction-metadata-and-seraph-governance-j4w7}
}

More from AI Collusion Research Sprint

All 392 entriesBrowse all projects

A Sprint is running now

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