Strategic Engineering Workshop
on LLMs and Game Theory

AAMAS 2026 | Paphos, Cyprus

https://cyprusconferences.org/aamas2026/workshops/

Latest update: Submit your paper through February 4th 6th!

OpenReview

Instructions

Template

TL;DR: Unlike traditional board games, real-world strategic interactions are messy. Traditional game theory thus needs a boost for the future of agentic AI. Come help us unlock truly strategic AI!

Keywords: Real-World Multi-Agent Scenario, LLM, Strategic World Model (e.g., Game Tree), Classical Game Solver, Real-World Decision

As AI agents diffuse into the fabric of our everyday lives, they will interact with each other (and also us) more and more often. Agents acting on behalf of humans might adopt their incentives, and so naturally some of these interactions will be cooperative, some competitive, and some mixed-motive. Game theory has long provided a framework for thinking about how to balance both individuals' and the collective’s incentives.

Real world scenarios include:

Negotiation (e.g., buying a car, new hire compensation, homework extension)
Zero-shot Coordination (e.g., answering questions with insufficient information and asymmetric goals (student-teacher), assisting a brainstorm with a new group)
Social Intelligence (e.g., offer your seat to the elderly, write a grateful / not indignant resignation letter)

Yet applying game theory to real world problems requires rigorous modelling and translation to precise mathematical specifications, algorithm selection and possibly novel design, scalable deployment, as well as potentially multiple rounds of iteration. In addition, the end product is often a limited, quantitative strategy specification lacking the typical richness or color of human interaction. We would like to bring together a community interested in reducing the friction in this pipeline. Moreover, modern AI agents should be able to complete this pipeline autonomously and transparently to provide the world with more strategic and interpretable agents.

This workshop will foster discussions around answering the following questions:

How can we build tools to automate the construction of strategic models (e.g., game trees) from rich multimodal inputs (e.g., textual/visual descriptions) as well as the exploration of alternative strategic models when new information arrives?
How can we evaluate the accuracy / usefulness of the constructed model in open ended domains?
How can we handle mismatches between players' different perceptions of strategic scenarios?
How can we collect data from humans to construct / verify the models? What form of interaction is most efficient / informative?
How might a strategic model be used to teach humans how to better interact and cooperate with each other?
Could building models help us further develop economic theory and design better institutions, markets and rules?

*While leveraging LLMs to construct strategic world models is the core focus, we will be accepting any works that look at the intersection between LLMs and Game Theory / Multi-Agent and are excited to foster wider discussions around this topic.

Imagine agents that can mediate human conflict and guide international diplomacy, agents that can assist us in navigating office politics, agents that fully autonomously re-negotiate your utility bill, and more. Come join us to usher in the next generation of strategic agents.

Call for Papers

We invite submissions exploring how large language models (LLMs) / foundation models (FMs) and game theory can enable strategic, interpretable AI agents for real-world scenarios.

The workshop is seeking submissions of research and industrial papers, including work on modelling, evaluation, algorithmic design, human data collection, and applications in negotiation, coordination, and everyday social intelligence, as well as demonstrations of agents succeeding (or failing) in strategic interactions.

Note: While the primary focus of the workshop is on leveraging LLMs to translate real-world scenarios to rigorous game-theoretic models, we will also consider papers that investigate other creative applications of LLMs to game theory or vice versa.

All deadlines are anywhere on earth (AoE) time.

Related Work / Inspiration:

Deng, Shilong, Yongzhao Wang, and Rahul Savani. "From Natural Language to Extensive-Form Game Representations." Proceedings of the 24th International Conference on Autonomous Agents and Multiagent Systems. 2025.
Xu, Zelai, Wanjun Gu, Chao Yu, Yi Wu, and Yu Wang. "Learning Strategic Language Agents in the Werewolf Game with Iterative Latent Space Policy Optimization." Forty-second International Conference on Machine Learning. 2025.
Mensfelt, Agnieszka, Kostas Stathis, and Vince Trencsenyi. "Autoformalization of game descriptions using large language models." arXiv preprint arXiv:2409.12300 (2024).
Gemp, Ian, Roma Patel, Yoram Bachrach, Marc Lanctot, Vibhavari Dasagi, Luke Marris, Georgios Piliouras, Siqi Liu, and Karl Tuyls. "Steering language models with game-theoretic solvers." In Agentic Markets Workshop at ICML 2024.
Daskalakis, Constantinos, Ian Gemp, Yanchen Jiang, Renato Paes Leme, Christos Papadimitriou, and Georgios Piliouras. "Charting the shapes of stories with game theory." In Creative AI Track at NeurIPS 2024.
Assos, Angelos, Idan Attias, Yuval Dagan, Constantinos Daskalakis, and Maxwell K. Fishelson. "Online learning and solving infinite games with an ERM oracle". In The Thirty Sixth Annual Conference on Learning Theory (pp. 274-324). PMLR. 2023.
Kroer, Christian, and Tuomas Sandholm. "A unified framework for extensive-form game abstraction with bounds." Advances in Neural Information Processing Systems 31 (2018).

Open Source Software (strategicwm):

To aid in exploring these ideas, we have also open sourced a python package strategicwm that presents one approach to generating game trees from natural language descriptions of multi-agent interactions. We will consider short papers that attempt to improve on this library and/or apply it to interesting multi-agent scenarios and provide insightful analysis. We encourage authors to think outside the box!

To use this library, you will need a (free tier) Gemini API key which can be obtained at aistudio.google.com (Get API key in the lower left panel, then Create API key in the upper right).

Submissions should adhere to the AAMAS proceedings format and fit into one of the following categories:

Short paper: A maximum of 4 pages, with no limit on references and appendices. These are aimed at highlighting successes / failures of existing approaches applied to open-ended multi-agent interactions.
Regular paper: A maximum of 8 pages, with no limit on references and appendices. These longer form submissions propose new techniques and solutions to automating or accelerating the construction of interpretable strategic world models and/or agents.

Submission Guidelines

Format your paper using the AAMAS 2026 paper template. Ensure the appendix is part of the same PDF as the main document.
Submit your paper via OpenReview.