SNEAK: Evaluating Strategic Communication and Information Leakage in Large Language Models

• URL: http://arxiv.org/abs/2603.29846v1
• Date: Tue, 31 Mar 2026 15:07:12 GMT
• Title: SNEAK: Evaluating Strategic Communication and Information Leakage in Large Language Models
• Abstract summary: We introduce SNEAK, a benchmark for evaluating selective information sharing in language models.<n>We evaluate generated messages using two simulated agents with different information states.<n>We show that strategic communication under asymmetric information remains a challenging capability for current systems.
• Score: 5.8120627413404184
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Large language models (LLMs) are increasingly deployed in multi-agent settings where communication must balance informativeness and secrecy. In such settings, an agent may need to signal information to collaborators while preventing an adversary from inferring sensitive details. However, existing LLM benchmarks primarily evaluate capabilities such as reasoning, factual knowledge, or instruction following, and do not directly measure strategic communication under asymmetric information. We introduce SNEAK (Secret-aware Natural language Evaluation for Adversarial Knowledge), a benchmark for evaluating selective information sharing in language models. In SNEAK, a model is given a semantic category, a candidate set of words, and a secret word, and must generate a message that indicates knowledge of the secret without revealing it too clearly. We evaluate generated messages using two simulated agents with different information states: an ally, who knows the secret and must identify the intended message, and a chameleon, who does not know the secret and attempts to infer it from the message. This yields two complementary metrics: utility, measuring how well the message communicates to collaborators, and leakage, measuring how much information it reveals to an adversary. Using this framework, we analyze the trade-off between informativeness and secrecy in modern language models and show that strategic communication under asymmetric information remains a challenging capability for current systems. Notably, human participants outperform all evaluated models by a large margin, achieving up to four times higher scores.

Related papers

• Multi-Agentic System Leveraging Open-Source LLMs to Mitigate Disinformation Threats [0.5629161809575014]
  This article proposes a novel approach based on a multi-agent system that emulates the decision-making processes of human annotators.<n>The proposed method achieves superior results compared to individual Large Language Models (LLMs), including GPT 4 and GPT 3.5.
  arXiv  Detail & Related papers  (2026-06-29T13:07:41Z)
• MT-PingEval: Evaluating Multi-Turn Collaboration with Private Information Games [70.37904949359938]
  We evaluate language models in multi-turn interactions using a suite of collaborative games that require effective communication about private information.<n>We find that language models are unable to use interactive collaboration to improve over the non-interactive baseline scenario.<n>We analyze the linguistic features of these dialogues, assessing the roles of sycophancy, information density, and discourse coherence.
  arXiv  Detail & Related papers  (2026-02-27T17:13:20Z)
• Verification Required: The Impact of Information Credibility on AI Persuasion [13.454393198058398]
  We introduce MixTalk, a strategic communication game for LLM-to-LLM interaction that models information credibility.<n>We evaluate state-of-the-art LLM agents in large-scale tournaments across three realistic deployment settings.<n>We propose Tournament Oracle Policy Distillation (TOPD), an offline method that distills tournament oracle policy from interaction logs and deploys it in-context at inference time.
  arXiv  Detail & Related papers  (2026-02-01T02:22:28Z)
• Evaluating Language Model Reasoning about Confidential Information [95.64687778185703]
  We study whether language models exhibit contextual robustness, or the capability to adhere to context-dependent safety specifications.<n>We develop a benchmark (PasswordEval) that measures whether language models can correctly determine when a user request is authorized.<n>We find that current open- and closed-source models struggle with this seemingly simple task, and that, perhaps surprisingly, reasoning capabilities do not generally improve performance.
  arXiv  Detail & Related papers  (2025-08-27T15:39:46Z)
• Teaching Language Models To Gather Information Proactively [53.85419549904644]
  Large language models (LLMs) are increasingly expected to function as collaborative partners.<n>In this work, we introduce a new task paradigm: proactive information gathering.<n>We design a scalable framework that generates partially specified, real-world tasks, masking key information.<n>Within this setup, our core innovation is a reinforcement finetuning strategy that rewards questions that elicit genuinely new, implicit user information.
  arXiv  Detail & Related papers  (2025-07-28T23:50:09Z)
• Towards eliciting latent knowledge from LLMs with mechanistic interpretability [1.3286418032136589]
  This work aims to explore the ability of current techniques to elicit hidden knowledge from language models.<n>We train a Taboo model: a language model that describes a specific secret word without explicitly stating it.<n>We develop largely automated strategies based on mechanistic interpretability techniques, including logit lens and sparse autoencoders.
  arXiv  Detail & Related papers  (2025-05-20T13:36:37Z)
• Remote Timing Attacks on Efficient Language Model Inference [63.79839291641793]
  We show it is possible to exploit timing differences to mount a timing attack. We show how it is possible to learn the topic of a user's conversation with 90%+ precision. An adversary can leverage a boosting attack to recover PII placed in messages for open source systems.
  arXiv  Detail & Related papers  (2024-10-22T16:51:36Z)
• KGV: Integrating Large Language Models with Knowledge Graphs for Cyber Threat Intelligence Credibility Assessment [38.312774244521]
  Cyber threat intelligence (CTI) is a crucial tool to prevent sophisticated, organized, and weaponized cyber attacks.<n>We propose Knowledge Graph-based Verifier (KGV), the first framework integrating large language models (LLMs) with simple structured knowledge graphs (KGs) for automated CTI credibility assessment.<n> Experimental results demonstrate that our KGV outperforms state-of-the-art fact reasoning methods on the CTI-200 dataset, achieving a 5.7% improvement in F1.
  arXiv  Detail & Related papers  (2024-08-15T11:32:46Z)
• Context-aware Communication for Multi-agent Reinforcement Learning [6.109127175562235]
  We develop a context-aware communication scheme for multi-agent reinforcement learning (MARL) In the first stage, agents exchange coarse representations in a broadcast fashion, providing context for the second stage. Following this, agents utilize attention mechanisms in the second stage to selectively generate messages personalized for the receivers. To evaluate the effectiveness of CACOM, we integrate it with both actor-critic and value-based MARL algorithms.
  arXiv  Detail & Related papers  (2023-12-25T03:33:08Z)
• Let Models Speak Ciphers: Multiagent Debate through Embeddings [84.20336971784495]
  We introduce CIPHER (Communicative Inter-Model Protocol Through Embedding Representation) to address this issue. By deviating from natural language, CIPHER offers an advantage of encoding a broader spectrum of information without any modification to the model weights. This showcases the superiority and robustness of embeddings as an alternative "language" for communication among LLMs.
  arXiv  Detail & Related papers  (2023-10-10T03:06:38Z)
• InfoBERT: Improving Robustness of Language Models from An Information Theoretic Perspective [84.78604733927887]
  Large-scale language models such as BERT have achieved state-of-the-art performance across a wide range of NLP tasks. Recent studies show that such BERT-based models are vulnerable facing the threats of textual adversarial attacks. We propose InfoBERT, a novel learning framework for robust fine-tuning of pre-trained language models.
  arXiv  Detail & Related papers  (2020-10-05T20:49:26Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.