Artificial intelligence (AI) tools are increasingly used for legal research, including by people navigating the justice system without a lawyer. They can help users explore legal questions, but they can also produce references to court decisions that do not exist. These fabricated citations can appear credible, making them difficult to identify without consulting the original decisions.
At the 2026 Law via the Internet (LVI) conference in Montreal, Tom Macintosh Zheng, co-founder of Courtready.ca, presented research on AI-generated legal errors in Canadian courts and tribunals, and the consequences for litigants and the justice system.
Fabricated Case Law
In his presentation, The Human Toll When AI Invents the Law, Macintosh Zheng examined cases in which AI-generated legal information had been submitted in court proceedings.
As of September 24, 2026, his research had identified 258 Canadian court and tribunal decisions involving AI-related misuse across 62 courts and tribunals. Of these, 215 involved citations to nonexistent cases. The number of documented decisions rose from eight in 2024 to 104 in 2025, with another 146 identified by September 2026.
These figures represent incidents documented in the research, rather than the total number of AI-generated errors in Canadian legal proceedings. They nevertheless show that fabricated legal authorities are appearing in cases before a range of courts and tribunals.
Self-Represented Litigants
One finding was particularly notable: 83% of the decisions examined involved self-represented litigants.
For people without legal training, AI tools can make legal information easier to explore. However, a citation that includes a plausible case name, court and reference number is not necessarily genuine.
Verifying a citation requires locating the original decision and confirming that it supports the legal argument being made. Someone unfamiliar with legal research may not know how to perform these checks. As a result, a person using AI to prepare a legal submission may unknowingly rely on a case that does not exist.
Judicial Costs
Fabricated citations also create additional work for courts. When legal submissions refer to nonexistent decisions, judges and court staff may need to investigate the references before assessing the arguments. Other parties can face additional work, and proceedings may be delayed.
Macintosh Zheng’s research also examined the consequences imposed in these cases. Among 250 proceedings reviewed, 77% resulted in either no recorded sanction or a warning.
The absence of a serious sanction does not mean that an error has no cost. Time spent checking fabricated authorities is time that cannot be devoted to other judicial work. The research draws attention to a consequence of AI-generated legal errors that is not always visible to the person using the tool: the additional burden placed on the justice system.
Source Verification
AI-generated legal references need to be checked against original sources. Courts, legal information institutes and other legal publishers make this possible by providing access to authentic legal materials.
Keeping legal information accessible, accurate, and up to date requires reliable publishing processes. Norma, Lexum’s legal information management and publishing platform, helps organizations manage, organize, and publish legal content online. While publishing platforms cannot prevent AI-generated errors, they make it possible for users to consult original legal sources and verify the information they receive.
As AI becomes more common in legal research, checking citations and legal arguments against published decisions remains an essential part of using these tools responsibly.
Reference
Tom Macintosh Zheng, The Human Toll When AI Invents the Law. Law via the Internet 2026, Montréal. Research available at Courtready.ca.
