Dec 12, 2025 | AI and product innovation
From Startup Vision to AI Breakthrough: One Year with Safe Sign
Last year, Thomson Reuters made its first pre-revenue acquisition in our 174-year history: Safe Sign, a startup focused on training large language models for the legal domain. This acquisition was more than a business decision. It was a conviction that the future of professional AI requires bringing together three rare ingredients that few organizations in the world possess: world-class expertise in training large-scale deep learning models, decades of highly curated authoritative content, and thousands of domain experts who understand what good looks like in high-stakes professional work.
Twelve months later, the collaboration between Thomson Reuters and Safe Sign continues to be one of the most exciting engines of innovation in our company. Together, we are building a foundation for the next era of professional-grade AI: one that combines frontier research, trusted data, and deep domain expertise.
Since joining Thomson Reuters, the Safe Sign team has become part of one integrated research organization within TR Labs, now known as Foundational Research. What began as a 15-person startup has grown into a 50-person global group that works seamlessly alongside our engineers, data scientists, and subject-matter experts across legal, tax, accounting, and compliance. Together, we are combining frontier research talent and deep domain expertise to accelerate development of the Thomson Reuters large language model, trained on professional-grade data and purpose-built for high-stakes work.
General-purpose models have unlocked incredible capabilities in language understanding and reasoning. But professionals work differently, and their tools must reflect that. Broad models are not optimized for the standards of accuracy, context, and accountability that legal and tax professionals require. They are powerful, but they are not precise.
A purpose-built model changes that. By training a purpose built LLM on decades of Thomson Reuters content and data, and by learning directly from thousands of subject-matter experts, our model can reason the way professionals do. It can connect facts with context, apply appropriate logic to specific scenarios, and provide grounded, verifiable output that supports confident decision-making. The result will be a model designed not for conversation, but for action and trust. A model that understands how real professional work gets done.
“This first year has shown what happens when world-class AI research meets trusted domain expertise,” said Dr. Jonathan Schwarz, co-founder of Safe Sign and now Head of Foundational Research within TR Labs. Before joining Thomson Reuters, Jonathan spent seven years as a senior researcher at Google DeepMind, where he helped pioneer methods in large-scale model training and neural optimization. “We are combining deep technical innovation with the quality and structure of Thomson Reuters data,” he said. “That partnership allows us to move faster, train smarter, and build something truly differentiated for the professional world.”
The collaboration extends beyond our own teams. Through the new Frontier AI Lab with Imperial College London, co-led by Jonathan, we are tackling some of the most important challenges in AI, including safer model training, efficient evaluation, and the long-term societal impact of advanced systems. This partnership reflects our belief that responsible innovation begins with scientific rigor and openness.
All of this work is paving the way for what comes next. In 2026, we intend to bring to market the first Thomson Reuters large language model designed from the ground up for professional-grade performance. It will combine our trusted content, unique data architecture, and agentic AI capabilities to deliver accuracy, transparency, and reliability at scale.
We are building a future where intelligence is not abstract and relegated to the written word but applied and embodied via AI. An industry first, where AI becomes the operating layer for professional work. And where innovation happens not just in research labs, but in the hands of professionals who depend on Thomson Reuters to move faster, think smarter, and act with confidence.
In just one year, this collaboration has shown what’s possible when bold vision meets real‑world scale. The year ahead will push the boundaries of professional‑grade AI and set a new benchmark for what the industry can achieve.
