Skip to content
Apart Research
Apart Research
Contact us
Sprint projectOct 26, 2026Jersey City

Letting AI Agents Choose Without Letting Them Whisper: Proved Leakage Caps Against Adaptive LLM Colluders

Teo Asinari

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

Read the report

Report: Letting AI Agents Choose Without Letting Them Whisper: Proved Leakage Caps Against Adaptive LLM Colluders

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

AI agents that work together can hide secret messages inside ordinary answers, for example in wording, formatting, or which of several equally good answers they pick. We study defenses that limit this. First, rewriting an agent’s message into a standard form removes every hidden signal we tried except one: which answer the agent chose. Second, we restrict that choice to a menu. A trusted model writes a short list of good answers and the agent may only pick one. Picking one of 8 options carries at most 3 bits, so at most 3 secret bits can leak per answer, and we prove this limit in Lean for every possible model. Third, networks built to depend only on approved information leak nothing, but leave the agent no freedom. Finally, we tested the defenses against real AI agents (GPT-4.1-mini, coached by Claude Sonnet 4.6) that improved their own secret code over 12 rounds, in five independent runs. Unprotected, they leaked up to 7 of 8 secret bits per answer. With menus of 2, 4, 8, 16 and 32 options, they leaked 1.0, 1.9, 2.9, 3.7 and 4.5 bits, against proven limits of 1 to 5, never meaningfully above them. Larger menus also gave better honest answers than free answers. The standard-form defense leaked nothing in three runs but about 2 bits in two, once the agents found its loophole.

Reviews

Judging this Sprint?

Review “Letting AI Agents Choose Without Letting Them Whisper: Proved Leakage Caps Against Adaptive LLM Colluders”

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.

No public critique yet.

Cite this project

@misc{asinari2026letting,
  title = {{Letting AI Agents Choose Without Letting Them Whisper: Proved Leakage Caps Against Adaptive LLM Colluders}},
  author = {Teo Asinari},
  year = {2026},
  month = oct,
  note = {Submitted to AI Collusion Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/letting-ai-agents-choose-without-letting-them-whisper-proved-leakage-caps-against-adaptive-llm-colluders-83e3}},
  url = {https://apartresearch.com/sprints/projects/letting-ai-agents-choose-without-letting-them-whisper-proved-leakage-caps-against-adaptive-llm-colluders-83e3}
}

More from AI Collusion Research Sprint

All 394 entriesBrowse all projects

A Sprint is running now

AI Collusion Research Sprint · Oct 23 - 25, 2026