AI4Law: AI for Law Workshop @ ICML 2026
📅 Friday, 10 July 2026 in 23 days
📍
Seoul, 
South Korea
A research workshop at ICML 2026 in Seoul (10 July 2026) on AI competence in law — legal reasoning, evaluation, and access to justice. Confirmed keynotes from Elliott Ash (ETH Zurich) and Peter Henderson (Princeton), with an extensive cross-institution panel.
AI for Law (AI4Law) is a research workshop at ICML 2026 in Seoul, South Korea, on 10 July 2026 (with the camera-ready/registration logistics tied to the main conference). It is organised around a single sharp question: what does it mean for an AI system to be competent in law, and how can such competence be built, evaluated and validated so that it generalises across legal systems while improving equitable access to justice?
The programme is structured around three themes. AI for Legal Reasoning covers long-form legal argumentation, applying law to facts, issue-spotting, rule interpretation, structured inference, tool use, and retrieval and supervision for legal tasks. AI Evaluation for Law focuses on measuring legal reasoning quality and doctrinal correctness, benchmarks and datasets for high-stakes settings, and cross-jurisdiction, multilingual evaluation. AI for Access to Justice addresses systems that help people understand and exercise their legal rights, human-AI collaboration in legal workflows, and the fairness and societal impact of legal AI.
The line-up is notably strong for a workshop. Confirmed keynote speakers are Elliott Ash (ETH Zurich) and Peter Henderson (Princeton), both leading figures in computational law and AI-and-law. The panel and organiser roster spans Stanford (Julian Nyarko; Robert Mahari of Stanford CodeX), Princeton, Peking University (Yansong Feng), the University of Hong Kong, Illinois Tech (Daniel Katz), Fordham, Hugging Face (Joel Niklaus), the Max Planck Institute and several European universities — a genuinely international, research-heavy group.
The workshop matters because legal reasoning is one of the most demanding tests of LLM reliability: it requires faithful application of rules to facts, doctrinal correctness and robustness to jurisdictional variation, exactly the properties general-purpose models still struggle with. Rigorous evaluation and a focus on access to justice push back usefully against hype. Submissions are double-blind via OpenReview (8 pages, ICML format). AI What's On lists it as a focused research workshop at a top-tier ML venue.
