# Do Latent Channels Actually Communicate? A Causal Audit of Latent Multi-Agent LLM

arXiv: https://arxiv.org/abs/2607.26773 · HTML: https://mlsanity.com/paper/2607.26773/

## Key facts

- Cited by 5 papers in the map (data through 14 Oct 2026).
- Placed in General LLM agent frameworks and benchmarks, one of 16 directions in LLM agents and tool use.
- Builds on 14 of its 22 references that are in the map.
- First posted on arXiv 29 Jul 2026.

## Placed in

- LLM agents and tool use / [General LLM agent frameworks and benchmarks](https://mlsanity.com/agents/general-llm-agent-frameworks-and-benchmarks/)

## Builds on (most cited first, 14 of 14)

- [Training Verifiers to Solve Math Word Problems](https://mlsanity.com/paper/2110.14168/) (2021)
- [Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challenge](https://mlsanity.com/paper/1803.05457/) (2018)
- [Let's Verify Step by Step](https://mlsanity.com/paper/2305.20050/) (2023)
- [Locating and Editing Factual Associations in GPT](https://mlsanity.com/paper/2202.05262/) (2022)
- [Training Large Language Models to Reason in a Continuous Latent Space](https://mlsanity.com/paper/2412.06769/) (2024)
- [On Variational Bounds of Mutual Information](https://mlsanity.com/paper/1905.06922/) (2019)
- [Towards Best Practices of Activation Patching in Language Models: Metrics and Methods](https://mlsanity.com/paper/2309.16042/) (2023)
- [Formal Limitations on the Measurement of Mutual Information](https://mlsanity.com/paper/1811.04251/) (2018)
- [Multi-Agent Cooperation and the Emergence of (Natural) Language](https://mlsanity.com/paper/1612.07182/) (2016)
- [Understanding the Limitations of Variational Mutual Information Estimators](https://mlsanity.com/paper/1910.06222/) (2019)
- [On the Pitfalls of Measuring Emergent Communication](https://mlsanity.com/paper/1903.05168/) (2019)
- [Let Models Speak Ciphers: Multiagent Debate through Embeddings](https://mlsanity.com/paper/2310.06272/) (2023)
- [Counterfactual Graph for Multi-Agent LLM Calibration](https://mlsanity.com/paper/2605.30653/) (2026)
- [When Less Latent Leads to Better Relay: Information-Preserving Compression for Latent Multi-Agent LLM Collaboration](https://mlsanity.com/paper/2604.13349/) (2026)

## Built on by (most cited first, 5 of 5)

- [When Does Latent Communication Pay? A Causal Audit of Relayed KV Caches in Multi-Agent LLMs](https://mlsanity.com/paper/2608.04893/) (2026)
- [What You Can't See Is What You Learn: Slot-Selective Evidence Masking Favors Compositional Generalization in Shared-Genome Language-Model Societies](https://mlsanity.com/paper/2608.20054/) (2026)
- [Portable Semantics, Private Dialects: Reuse and Negative Transfer in Latent Communication Between Language-Model Cells](https://mlsanity.com/paper/2609.11365/) (2026)
- [Where Should Language Sit in a Multimodal Model? Lessons from What Language Does to Human Perception and Cognition](https://mlsanity.com/paper/2609.07474/) (2026)
- [Social Circuits behind Multi-agent Echo Chambers](https://mlsanity.com/paper/2609.34444/) (2026)

## Abstract (arXiv)

Latent communication in large language model (LLM)-based multi-agent systems (MAS) transmits continuous internal representations instead of text, but greater representational capacity does not establish that the receiver uses task-relevant information. End-task performance alone also cannot reveal whether an observed effect depends on message presence, content generated for the evaluated example, or information supplied by a separate agent. We introduce a causal audit that applies controlled message replacements at the boundary where the sender-produced representation enters the receiver. Four message settings support five measurements of encoded sender information, receiver sensitivity to message presence and identity, the task value of example-specific content, and the additional value supplied by a separate agent. We apply the audit to latent relay with Qwen3-4B and Qwen3-8B on GSM8K, ARC-C, and MATH-500. On GSM8K, the Qwen3-4B overall performance effect of -1.00 percentage point decomposes into a -6.17-point effect retained by an other-example message and a +5.17-point effect attributable to example-specific content; both component directions reverse at 8B. On MATH-500, the Qwen3-4B gain of 15.00 points comprises 8.33 points retained by an other-example message and 6.67 points attributable to example-specific content, while the 8B gain is dominated by the former component. Self-substitution comparisons further show that example-specific content and other-agent value are distinct. These results show that aggregate accuracy does not identify how a latent message affects the receiver and motivate controlled message comparisons as a standard evaluation for latent communication.