Thomson Reuters launches proprietary AI model for legal work

Thomson Reuters is taking a different path from the biggest AI labs by focusing exclusively on legal expertise. The company says Thomson needs to set the frontier of intelligence for legal, which is a different job than what frontier labs are doing. Instead of building a foundation model from scratch, it began with an open-weight model and added its proprietary content, training methods and professional expertise, which reduced both training and inference costs.

The training process included realigning the base model with Thomson Reuters' values, pretraining on the company's content, and reinforcement learning that taught the model to work with tools such as Westlaw and Practical Law. Westlaw encompasses over 40,000 individual databases and more than 150 years of legal publishing. Hundreds of subject-matter experts helped define training objectives and judge responses. The research team focused on continual learning to add domain skills without erasing existing capabilities.

Internal tests showed Thomson is broadly competitive with leading models when all had access only to the web, and moved to roughly equal or slightly better performance when connected to Thomson Reuters content. These results have not yet received extensive independent validation. Only about 10% of the company's total information base has been used so far, and the next step is turning the most useful content into better training signals. Hron acknowledged that maintaining a proprietary model raises questions about keeping pace with faster-moving AI laboratories, but argued improvements in open models will give stronger foundations for later versions.