LawThinker: A Deep Research Legal Agent in Dynamic Environments
Xinyu Yang, Chenlong Deng, Tongyu Wen, Binyu Xie, Zhicheng Dou
2026-02-13
Summary
This paper introduces LawThinker, a new computer program designed to help with legal reasoning, aiming to make sure not just the final answer is correct, but also that the steps taken to get there follow legal procedures.
What's the problem?
Current computer programs that try to do legal reasoning often make mistakes not just in the final conclusion, but also in the steps they take to reach it. For example, they might cite a law that doesn't actually apply to the case. These errors can build up and lead to a completely wrong outcome, and existing systems don't have a good way to catch these problems as they happen during the reasoning process.
What's the solution?
LawThinker tackles this by using a strategy of 'Explore-Verify-Memorize'. After every step where it looks up information, it immediately checks that information for accuracy, relevance to the case, and whether it follows proper legal procedures. This checking is done by a 'DeepVerifier' module. It also has a 'memory' to remember what it's learned in previous steps, which is helpful for complex cases. Essentially, it double-checks itself constantly.
Why it matters?
LawThinker is important because it significantly improves the reliability of computer-assisted legal reasoning. It's shown to be much better than existing methods at getting the right answer *and* following the correct legal process, which is crucial in law. This means it could be a valuable tool for lawyers and legal researchers, and it shows a step forward in building trustworthy AI for the legal field.
Abstract
Legal reasoning requires not only correct outcomes but also procedurally compliant reasoning processes. However, existing methods lack mechanisms to verify intermediate reasoning steps, allowing errors such as inapplicable statute citations to propagate undetected through the reasoning chain. To address this, we propose LawThinker, an autonomous legal research agent that adopts an Explore-Verify-Memorize strategy for dynamic judicial environments. The core idea is to enforce verification as an atomic operation after every knowledge exploration step. A DeepVerifier module examines each retrieval result along three dimensions of knowledge accuracy, fact-law relevance, and procedural compliance, with a memory module for cross-round knowledge reuse in long-horizon tasks. Experiments on the dynamic benchmark J1-EVAL show that LawThinker achieves a 24% improvement over direct reasoning and an 11% gain over workflow-based methods, with particularly strong improvements on process-oriented metrics. Evaluations on three static benchmarks further confirm its generalization capability. The code is available at https://github.com/yxy-919/LawThinker-agent .