SocialVeil: Probing Social Intelligence of Language Agents under Communication Barriers

ICLR 2026 Conference Submission15067 Authors

Keywords: Social Interaction, Agent, Social intelligence, Alignment, Large Language Models, Evaluation

Abstract: Large language models (LLMs) are increasingly evaluated in interactive environments to test their social intelligence. However, existing benchmarks usually assume idealized communication between agents, limiting our ability to diagnose whether LLMs can maintain and repair interactions under real-world conditions. To close this gap, we present SocialVeil, a social learning environment that can simulate social interaction under cognitive-difference-induced communication barriers. SocialVeil introduces three representative types of such disruption, semantic vagueness, sociocultural mismatch, and emotional interference. SocialVeil also introduces barrier-aware evaluation metrics, unresolved confusion and mutual understanding, which complement standard goal-oriented evaluation by assessing agents' capability of maintaining interaction in impaired communication. Experiments across 720 scenarios and four frontier LLMs show that barriers consistently impair performance, with mutual understanding reduced by over 45% on average, and confusion elevated by nearly 50\%. Human evaluations validate the fidelity of these simulated barriers (ICC$\approx$0.78, Pearson r$\approx$0.80). We further demonstrate that adaptation strategies (Repair Instruction and Interactive learning) only have a modest effect that remains far from barrier-free performance. This work takes a step toward bringing social interaction environments closer to real-world communication, opening broader opportunities for exploring the social intelligence of LLM agents.

Primary Area: alignment, fairness, safety, privacy, and societal considerations

Submission Number: 15067