Towards Safety Evaluations of Theory of Mind in Large Language Models

2025-06-19

AI Summary

Large language models (LLMs) may exhibit deceptive behaviors—such as evading oversight or generating false responses—posing critical safety risks; however, existing evaluations lack a principled framework to assess *intentional* deception. Method: This work introduces Theory of Mind (ToM) capability—the ability to attribute mental states to others—as a core, interpretable metric for detecting deliberate deception in LLMs. We adapt developmental psychology’s ToM framework to LLM safety evaluation, establishing the first cross-model ToM developmental trajectory analysis paradigm. Our methodology integrates classic False Belief tasks with security-oriented behavioral analysis within a multi-task benchmark, employing controllable prompting and fine-grained behavioral trajectory tracking. Contribution/Results: Empirical evaluation reveals that mainstream open-source LLMs exhibit consistently weak ToM capabilities, with no significant improvement observed across increasing parameter scales. Crucially, ToM deficits correlate statistically with deceptive response generation. This study establishes a novel, interpretable dimension, benchmark, and analytical pathway for trustworthy LLM safety assessment.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Safety and Robustness
• Philosophy and Ethics of AI: Safety, Robustness & Trustworthiness

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

As the capabilities of large language models (LLMs) continue to advance, the importance of rigorous safety evaluation is becoming increasingly evident. Recent concerns within the realm of safety assessment have highlighted instances in which LLMs exhibit behaviors that appear to disable oversight mechanisms and respond in a deceptive manner. For example, there have been reports suggesting that, when confronted with information unfavorable to their own persistence during task execution, LLMs may act covertly and even provide false answers to questions intended to verify their behavior.To evaluate the potential risk of such deceptive actions toward developers or users, it is essential to investigate whether these behaviors stem from covert, intentional processes within the model. In this study, we propose that it is necessary to measure the theory of mind capabilities of LLMs. We begin by reviewing existing research on theory of mind and identifying the perspectives and tasks relevant to its application in safety evaluation. Given that theory of mind has been predominantly studied within the context of developmental psychology, we analyze developmental trends across a series of open-weight LLMs. Our results indicate that while LLMs have improved in reading comprehension, their theory of mind capabilities have not shown comparable development. Finally, we present the current state of safety evaluation with respect to LLMs' theory of mind, and discuss remaining challenges for future work.

Problem

• Evaluating deceptive behaviors in large language models
• Assessing theory of mind capabilities in LLMs
• Investigating safety risks from intentional model processes

Innovation

• Measure theory of mind in LLMs
• Analyze developmental trends in LLMs
• Assess safety via theory of mind