Professor Samuel Dahan has received $498,885 from the Natural Sciences and Engineering Research Council of Canada (NSERC) for a global, interdisciplinary research initiative set to transform how courts use artificial intelligence. The project will tackle one of the most urgent challenges facing justice systems worldwide: how to deliver fair, consistent, and timely decisions in an era of growing demand and technological disruption.
Dahan is Director of the Conflict Analytics Lab, which, in 2023, launched OpenJustice — a pioneering open-source approach that enables generative AI to perform legal tasks through expert-designed reasoning models. As principal investigator for this new research project — titled “Courtroom Intelligence: Co Designing AI for Courts” — he’ll lead the development of an open-source “judicial AI” framework designed specifically for courts. The project will be funded through NSERC’s Discovery Horizons program over five years.
The project, funded through NSERC’s Discovery Horizons program over five years, is part of more than $14 million NSERC awarded to Queen’s researchers in its 2026 Discovery Research Program.
A global, interdisciplinary collaboration
The initiative brings together experts in law, computer science, linguistics, and social science, along with leading judicial partners. In addition to Queen’s Smith School of Business and Queen’s Department of Electrical and Computer Engineering, partnering academic institutions include Toronto Metropolitan University, McGill University, Harvard Law School, University College London, and Leiden University / Council of Europe. Judicial and institutional collaborators include former Supreme Court of Canada Justice Thomas Cromwell, the French Cour de cassation, and the Hong Kong Judiciary.
“This project is unique because it’s truly co-designed with courts,” Dahan says. “We are not building technology in isolation; we are working directly with judges, clerks, and legal practitioners to ensure it meets real-world needs.”
Addressing the access-to-justice crisis in an age of unreliable AI
The initiative responds to a stark reality: nearly five billion people worldwide lack access to adequate legal assistance. In Canada alone, 78 per cent of individuals facing serious legal issues do not receive help. Courts are increasingly overwhelmed, particularly by rising numbers of self-represented litigants who must navigate complex procedures without legal counsel.
Artificial intelligence has often been promoted as a solution, but existing legal AI tools have proven unreliable in practice. Studies show that current systems frequently generate incorrect or fabricated legal information — so-called “hallucinations” — and may even reinforce users’ misconceptions. In court settings, this can result in flawed filings, wasted judicial resources, and delays in proceedings.
“AI has enormous potential to support access to justice,” Dahan says. “But today’s systems are not designed for the realities of legal reasoning or judicial work. Our goal is to build AI systems whose reasoning courts can inspect, evaluate, and trust — systems grounded in how legal decisions are made, not just in large volumes of text.”
Moving beyond data: modelling legal reasoning
A central insight of the project is that law cannot be reduced to datasets alone. While most current legal AI systems rely on large databases of statutes and case law, they fail to capture the experiential and contextual knowledge that underpins real-world decision-making. In fact, more than 90 per cent of legal disputes are resolved outside formal court rulings, meaning that critical knowledge is never recorded in legal texts.
To address this gap, OpenJustice introduces a new approach. Instead of training AI solely on legal data, it models the reasoning processes used by judges and lawyers. At the heart of this innovation is a “no-code” framework called reasoning flows, which allow legal professionals to build step-by-step representations of legal analysis — such as identifying relevant statutes, evaluating evidence, and applying precedent — using a simple visual interface. Each step is transparent, verifiable, and grounded in expert knowledge.
“Rather than treating AI as a black box, we enable judges and lawyers to shape how it thinks,” Dahan explains. “They can encode their reasoning directly into the system, ensuring that outputs are explainable, auditable, and aligned with legal standards.”
Early pilot studies suggest that this approach significantly reduces errors compared to traditional AI systems, while improving transparency and user confidence.
Two key areas of innovation for courts
The project will focus on two primary areas where AI can deliver immediate benefits to judicial institutions:
1. Drafting assistance and decision standardization
Judicial decisions must be clear, consistent, and accessible, but courts have largely been left behind in adopting modern drafting technologies.
OpenJustice will provide AI-assisted drafting tools tailored for judges, allowing them to generate structured, standardized reasoning while preserving legal nuance. These tools are designed to improve clarity, reduce workload, and ensure consistency across decisions, particularly in high-volume court systems.
Unlike generic writing AI, the system embeds legal reasoning directly into drafting templates, ensuring outputs meet the precision and neutrality required in judicial writing.
2. Evidence admissibility and deepfake detection
The rise of digital evidence, including AI-generated “deepfake” media, poses a growing challenge for courts. Existing legal frameworks often lack the tools needed to verify authenticity, leaving judges to rely on costly expert testimony or ad hoc solutions.
The project will explore and evaluate methods for assisting courts in assessing synthetic and manipulated evidence. Using multimodal analysis — examining visual, audio, and metadata signals — the platform will help courts identify manipulated evidence and assess whether it meets legal standards.
By integrating technological detection with legal reasoning, the system aims to provide a transparent and standardized approach to evaluating digital evidence.
Open-source approach to public trust
A defining feature of OpenJustice is its commitment to open-source design. Unlike proprietary legal AI systems developed by private companies, the platform will be accessible to courts, legal-aid organizations, and researchers worldwide. This approach is intended to prevent vendor lock-in, promote transparency, and enable public oversight of how AI is used in the justice system.
Unlike many closed systems, open-weight models can be deployed locally within court infrastructure, reducing reliance on external providers and giving institutions greater control over sensitive judicial data. Recent research from the Conflict Analytics Lab also suggests that combining smaller open-weight models with structured legal reasoning can achieve performance comparable to, and in some cases exceeding, much larger frontier AI systems, while operating at a fraction of the cost.
“Justice is a public good,” Dahan emphasizes. “The tools that support it should be open, accountable, and subject to scrutiny.”
Training the next generation of legal AI experts
The project will also serve as a training ground for students and researchers at the intersection of law and technology. During each of the project’s five years, more than 20 trainees — from undergraduate students to postdoctoral fellows at collaborating academic institutions — will participate in developing and testing the platform.
Toward more efficient, transparent justice
Ensuring that similar cases are treated similarly is fundamental to the rule of law. Yet courts often struggle with inconsistencies across decisions, known as “doctrinal drift.” The OpenJustice platform will include tools to analyze judicial reasoning across cases, identify inconsistencies, and highlight where precedents are being applied unevenly. By combining statistical analysis with reasoning-based AI, the system aims to support more coherent and predictable decision-making.
Over its five years, the project will move from design and pilot testing to full evaluation and knowledge sharing. Expected outcomes include new benchmarks for judicial AI, policy guidance for courts and regulators, and a validated open-source platform ready for broader adoption.
“In the long run, this is about strengthening public confidence in the justice system,” Dahan says. “If we can build AI that reflects how judges actually reason — and make that process visible — we can make justice more reliable for everyone.”
Expanding the platform, transforming the future
This NSERC funding builds on more than $2 million in prior funding from NSERC Alliance, Mitacs, the Social Sciences and Humanities Research Council (SSHRC), the Law Foundation of Ontario, and project partners to expand the OpenJustice platform. It also marks a step toward a larger goal.
“The Conflict Analytics Lab is preparing a 2027 CFI Innovation Fund proposal for a roughly $3.9-million project to build dedicated, sovereign computing and secure data storage for legal AI — the first of its kind in Canada and among the first in the world,” Dahan explains.
“The aim is to let courts, legal-aid organizations, and law firms develop and run open-weight legal models locally, on Canadian infrastructure, rather than relying on external providers.”