Thomson Reuters Taps Decades of Content to Train AI

Thomson Reuters unveiled its in-house large language model, named "Thomson," on August 24th. The AI model was trained on the company's vast repository of content spanning Westlaw, Practical Law, Checkpoint, and Reuters sources over the years. Created from a foundation of open-source models, Thomson leverages the company's proprietary expertise instead of relying on customer data or incurring inference costs associated with leading models.

Designed with legal professionals in mind, Thomson ensures accountability by not utilizing customer data without explicit consent. CEO Steve Hasker highlighted that Thomson represents a competitive edge for Thomson Reuters, as it combines decades of proprietary content with editorial prowess. The AI platform is initially being utilized in CoCounsel Legal, with plans to expand its capabilities across the legal and tax domains.

Thomson's development challenges the notion that larger, more computationally intensive models are always superior. Chief Technology Officer Joel Hron emphasized that a strong foundation, deep specialization, and control over the AI system can yield highly capable, efficient intelligence. Hasker announced in March 2024 that Thomson Reuters had allocated $8 billion for AI investments, signaling a transition from a content provider to an AI-driven professional services company.

As of February 2025, CoCounsel, Thomson Reuters' AI solution for automating complex workflows, had already been adopted by 1 million professionals across 107 countries.