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

Cutting the Wire: Countermeasures Against Learned Vector Channels Between Language Models

jiho yun · Team groothaha

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

Read the report

Report: Cutting the Wire: Countermeasures Against Learned Vector Channels Between Language Models

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

Two frozen LLMs can learn to communicate through learned vectors instead of text: an adapter pools the sender's hidden states into 16 vectors the receiver reads in place of a message, reaching 94-96% accuracy on a recall task against 29-37% for 16 tokens of text - and a text monitor sees none of it. We take the defender's side. Snapping each incoming vector to its nearest token embedding removes all 49 channels we trained, into three receiver models, and leaves text unchanged. A linear probe trained on 2,000 paired examples recovers 80-88% of transmitted digits (chance 10%). Because defences face adaptive attackers, we then fine-tune channels against each deployed defence - and each evasion walks into another defence: a channel trained to evade the probe succeeds but stays far from the vocabulary, where a paired-data-free detector flags it; a channel trained to mimic tokens gets close enough that snapping still removes it; and a channel trained through the snap itself collapses in all 3 seeds. The channel is shared across three models and holds about twice text accuracy up to 395-token messages. Together, the defences corner an adaptive attacker: whatever it evades, the remaining message is either off-manifold and detectable or on-vocabulary and canonicalisable.

Reviews

Judging this Sprint?

Review “Cutting the Wire: Countermeasures Against Learned Vector Channels Between Language Models”

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. You picked a real gap: defending against learned vector channels between agents, as opposed to building them. The controls are good. You ran swapped, zero and no-message baselines, applied every countermeasure to text as well, and gave the probe a shuffled-label control. The headline defence is weaker than the framing suggests. A fully snapped message is 16 real tokens, so removing the advantage over text follows mostly from how the defence is built. It also gives up all of the vector bandwidth. Your snap-trained attacker scores 0.000, which is below the 0.373 that plain text gets through the same path. That points to an optimiser failure, not a strong defence, so "cornered" overstates it. Next steps: train one attacker against the full stack together, push mimicry past λ=3 (it reached cos 0.931 against your 0.95 threshold), report the detector's false-positive rate on benign vector traffic, and use a task where the agents have a reason to collude.

    Read full reviewShow less

  2. I like your idea for the covert communication channel. The defense you present is true almost by construction - and it feels more like an issue of the way your training setup embedded the vector channel then a robust defense mechanism.

  3. Good direction and experimental structure. Keep in mind that disabling a communication mechanism is not the same as securing it. For example, preserving legitimate latent communication is valuable (current approach removes the capability entirely). Would recommend pushing in the direction of selective detection/mitigation of malicious behavior with next experiments.

Cite this project

@misc{yun2026cutting,
  title = {{Cutting the Wire: Countermeasures Against Learned Vector Channels Between Language Models}},
  author = {jiho yun},
  year = {2026},
  month = oct,
  note = {Submitted to AI Collusion Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/cutting-the-wire-countermeasures-against-learned-vector-channels-between-language-models-gre5}},
  url = {https://apartresearch.com/sprints/projects/cutting-the-wire-countermeasures-against-learned-vector-channels-between-language-models-gre5}
}

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