The 2nd Workshop on LLM Agents
for Social Simulation

Overview

Social simulation plays an important role in domains where human studies face practical constraints, such as sociology, psychology, medicine, and education. Conventional social simulations often relied on rule-based, statistical, or mechanistic models with predefined assumptions, which can oversimplify the complexity, diversity, and context dependence of real-world human behavior.

Recent advances in Large Language Models (LLMs) offer a promising opportunity by enabling agents to interpret contextual information, perform flexible reasoning, and adapt their behaviors to evolving environments, thereby opening new frontiers for modeling human behavior, social interactions, and collective dynamics in support of controlled analysis, counterfactual reasoning, and decision support.

Despite these opportunities, developing LLM-agent-based social simulations also raises significant challenges — including how to model personas and domain knowledge; how to evaluate and analyze simulated behaviors; how to mitigate risks and ensure safety, ethics, and accountability, especially in error-sensitive domains; and how to build scalable infrastructure for large-scale and long-horizon simulations.

Themed "Fundamental Challenges and Real-World Applications", the second LASS workshop aims to bring together researchers and practitioners from diverse disciplines to explore how LLM agents can be more faithfully modeled, rigorously evaluated, safely governed, and effectively applied to address real-world problems in vertical domains — with careful consideration of their unique requirements, challenges, and practical limitations, thereby promoting human well-being and broader societal benefit.

Building on the success of LASS 2025 (CIKM 2025, Seoul) — which attracted around 25 participants, received 15 valid submissions (each with 2–3 reviews), accepted 8 papers, presented 3 LASS Outstanding Paper Awards, and recommended 3 papers for fast-track consideration to JCST — this second edition will further emphasize real-world vertical-domain applications, with stronger attention to expert knowledge, domain-specific safety and ethical requirements, and pathways toward responsible real-world deployment.

Call for Papers

By fostering dialogue among AI, NLP, HCI, and a broad range of vertical domains, LASS aims to advance LLM-agent-based social simulations that are technically faithful, safe, ethical, and responsive to the unique requirements of vertical domains. We welcome original submissions that contribute to one or more of the following thematic axes:

Program

Half-day workshop · November 8, 2026 · Rome, Italy. The schedule is preliminary and subject to adjustment.

Program Committee

Each submission is assigned to at least three PC members without conflicts of interest, evaluated based on relevance, novelty, technical quality, and clarity. Additional reviewers may be invited according to the number of submissions.