This comprehensive survey examines how Large Language Models (LLMs) develop Theory of Mind (ToM) - the ability to attribute mental states to others and predict behavior.

• LLMs demonstrate measurable Theory of Mind capabilities through behavioral evaluations and neural representations
• Advanced ToM in AI systems creates significant security vulnerabilities including manipulation, deception and exploitation
• Current evaluation frameworks are insufficient for measuring sophisticated ToM capabilities
• Research priorities should include developing robust ToM assessment methods and implementing safety mitigations

Why it matters: As LLMs develop more sophisticated social intelligence, understanding their ability to model human minds becomes critical for anticipating security risks and building safer AI systems.

A Survey of Theory of Mind in Large Language Models: Evaluations, Representations, and Safety Risks