UBC & Caseway team up on court‑trained legal AI system

The University of British Columbia and Vancouver-based startup Caseway have launched a two-year research collaboration to develop a legal AI system trained only on court decisions.

The partners said the project focuses on reliability in legal research, after a series of public examples of generative AI producing false citations and invented precedents. They plan to build a model that draws from primary legal sources rather than scraped internet content.

The work brings together Dr. Vered Shwartz, an Assistant Professor of Computer Science at UBC and Canada CIFAR AI Chair, and Alistair Vigier, CEO and Co-Founder of Caseway. The project has backing from research grants from NSERC and Mitacs. It will fund a postdoctoral researcher and a graduate student.

Caseway said it already indexes millions of court decisions. The partners said their approach centres on grounding answers in case law and improving the way the system retrieves relevant decisions. They also said they will test the system rigorously and explore methods that combine large language models with symbolic reasoning.

Policy backdrop

The initiative comes amid shifting expectations for general-purpose AI systems in sensitive areas such as law. The release cited a policy change that restricted legal advice from a leading chatbot. It also linked the changes to rising use of AI tools by members of the public who cannot afford legal services.

Pricing pressure has pushed some people towards online research and automated tools. Caseway referenced typical hourly rates for lawyers and said many people already rely on online searches, which can return misleading information.

"Access to justice is a huge problem, here in Canada, roughly 70% of people in the justice system lack legal representation," said Caseway's Alistair Vigier.

"With ChatGPT pulling back, that's one less tool people might have experimented with for help. We see our legal AI as a way to level the playing field. Whether you're a lawyer at a law firm or someone preparing for court on your own, this technology will offer reliable legal research at your fingertips, so you're not left in the dark," said Vigier.

Hallucination risk

Researchers and courts have flagged the risk of AI systems generating plausible-sounding legal references that do not exist. The UBC and Caseway team said they want to reduce these errors by limiting training and retrieval to court decisions and by providing source attribution.

Shwartz said current large language models show traits that make them unsuitable for uncritical use in law.

"LLMs are already being used in fields such as law, but they still suffer from limitations like overconfidence and the tendency to hallucinate facts," Dr. Vered Shwartz noted, citing her concern that rushing AI into legal use without fixes could be harmful.

The team said it will focus on retrieval quality and how the system links responses to relevant decisions. They described a plan that includes evaluation of outputs and testing of the tool as it develops.

"People are already using LLMs to get answers for legal questions. Our goal is to deliver these answers in the context of Canadian law, with source attribution and reducing hallucinations. By focusing on tasks such as improving the granularity of the retrieval of case laws, combining LLMs with symbolic reasoning, and rigorously testing the tool, we aim to create an AI that professionals and the public can trust to provide accurate legal information," said Shwartz.

Open-source base

The partners said they will build the tool on open-source AI foundations. They linked that choice to transparency and to the ability to customise the model for legal tasks.

Vigier framed the project as a turning point for the company's work in legal technology.

"This partnership is a game-changer for legal technology," said Alistair Vigier, CEO of Caseway.

"Combining Caseway's practical innovation with UBC's world-class AI research, we're poised to create something truly transformative. We want to make sure anyone, from lawyers to self-represented people, can get reliable legal answers when they need them," said Vigier.

Access focus

Caseway said it has discussed legal information needs with groups including the Indigenous Bar Association. The partners said cultural and linguistic nuances will form part of the responsible AI work around the model.

"We want to ensure this technology benefits all Canadians, including those who have historically faced barriers in the legal system," said Vigier.

Caseway also outlined plans for a shift to a freemium model. It said it currently charges for subscriptions and seeks grants and public funding for free access to core legal research features. The company said it does not use traditional venture capital and relies on angel investment and government grants.

"Our vision is to provide the average person the same research power that lawyers have, without the hefty price tag," said Vigier, noting that sustainable grant support will enable the company to keep the lights on while offering a basic version of the tool to all.

The partners said the research will run over two years, with work starting at UBC in Vancouver and alongside Caseway's data engineering team.