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

The Truthfulness Trap: Why Claim Verification Missed Multi-Agent Steering

Dele Atanda · Team metaProof///Metaye

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

Read the report

Report: The Truthfulness Trap: Why Claim Verification Missed Multi-Agent Steering

Open in a new tab (opens in new tab)
More on dev-beta.aigentz.me (opens in new tab)
Share

We built a deterministic claim-attestation enforcer for Colosseum’s JiraTicket environment. It passed a blind held-out parser gate: 103/103 precision, 0/46 false self-attributions, 0/120 false denials. Neither positive control produced reliable harm. Qwen2.5-32B coalitions coordinated cooperatively (honest regret 0). An exploratory steering effect (regret 22.2 vs 0.0) did not replicate on fresh seeds (0/3). So the enforcement comparison was not run, and we make no harm-reduction claim. An offline replay found that none of the 17 heuristic-identified steering messages contained a parseable claim, which puts all of them outside the enforcer’s inspection boundary. Our conclusion is that the truthfulness of the information an agent passes on is distinct from the legitimacy of its influence over another agent’s decision.

Reviews

Judging this Sprint?

Review “The Truthfulness Trap: Why Claim Verification Missed Multi-Agent Steering”

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 builds a fact checker for the Colosseum Jira task which verifies the agent's claims about their own cost and can block the false ones. It relates to Track 3 with the question of whether such constitutional constraints reduce collusive harm without killing legit cooperation. Execution discipline is impressive- every run was bound to criteria committed before it ran, the parser was tested on held-out sets, and the author stopped when the positive controls failed instead of running the main test anyway. The most useful finding is that none of the 17 steering messages contained a checkable claim, so a fact checker would never have touched them.

    The honest headline is that the primary experiment never ran because no positive control held. The coalition coordinated without reliably harming anyone and the steering effect didn’t replicate on fresh seeds. So there is no result on whether enforcement reduces harm. Samples are very small, and the parser’s test sets were AI generated and AI labelled. Most of the harness seems to have been built before the sprint, and since the repo is not public, the hash-bound verification the paper describes can not be checked by judges. The later sections on Invariant Intelligence and Effective Capability are interesting, but they aren’t tested here and add length to the presentation without evidence.

    I would suggest first building a setup where colluders reliably harm the honest agent, for example by giving them more reason to steer or by limiting what the honest agent knows about costs. Then test a control on who can direct another agent’s choice, which the paper itself points to as the real gap. Making the repo public would also let others check the work.

    Read full reviewShow less

Cite this project

@misc{atanda2026truthfulness,
  title = {{The Truthfulness Trap: Why Claim Verification Missed Multi-Agent Steering}},
  author = {Dele Atanda},
  year = {2026},
  month = oct,
  note = {Submitted to AI Collusion Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/the-truthfulness-trap-why-claim-verification-missed-multiagent-steering-i9hl}},
  url = {https://apartresearch.com/sprints/projects/the-truthfulness-trap-why-claim-verification-missed-multiagent-steering-i9hl}
}

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