AI-Hallucinated Case Law in Canada: What We Found, What It Means, and What Needs to Change

Between January 2024 and March 10, 2026, Canadian courts caught at least 211 non-existent cases cited as real law across 111 decisions. The count is rising, and self-represented litigants are the most exposed.

Courtready research illustration on AI-hallucinated case law in Canadian courts

We spent the past year searching for AI-hallucinated cases submitted as real law in Canadian courts. We found too many. Between January 2024 and March 10, 2026, courts and tribunals caught parties citing at least 211 cases that do not exist, across 111 separate decisions and 42 courts and tribunals nationwide. In 82 of those 111 decisions, the court found or presumed that AI tools generated the fictitious cases.

This is no longer an isolated problem. It is systemic, and it is accelerating.

The figures are conservative. They count only what judges and adjudicators caught, remarked on, and published. In nearly half the decisions, 54 of 111, the court flagged fictitious citations without naming the specific cases, so the true count per decision is often unknown. Citations that slipped past the bench, past opposing counsel, and past the parties never enter the data at all. The 211 figure is a floor. The ceiling is anyone’s guess. The full record lives in our live database of fictitious citations in Canadian courts, updated weekly.

The trajectory is steep

Decisions flagging fictitious citations
Figure 1. Canadian decisions flagging fictitious citations, by year. The 2026 figure covers only the first ten weeks (to March 10, 2026).

We identified 7 such decisions in 2024. That rose to 80 in 2025. In just the first ten weeks of 2026, we found 24 more. At that pace, the 2026 total is on track to exceed the year before it, with most of the year still to come.

The increase tracks two trends moving in the same direction. Generative AI tools are becoming more accessible, and the number of Canadians appearing in court without a lawyer keeps growing. Together, they push more unverified citations toward the bench every month.

Self-represented litigants are the most exposed

87 of the 111 decisions involved fictitious cases that appeared in the submissions of self-represented litigants.

If you are representing yourself and using AI to help with legal research, this matters to you directly. AI tools do not just occasionally make mistakes. They fabricate entire cases. They will hand you a case name, a citation, a year, and a court, and all of it will look real. When the judge searches for that case and finds nothing, the consequence falls on you.

Legal professionals have submitted fictitious cases too, so the problem is not confined to any one group. But self-represented litigants are the most exposed, and many may not even know that AI tools tend to hallucinate. Many turn to these tools for the same reason they represent themselves: they cannot afford a lawyer. For someone navigating an unfamiliar system without guidance, there is no obvious signal that anything has gone wrong.

What needs to change

We built CaseCheck to address part of this problem. It lets users upload a list of cases they plan to cite, extracts each citation, and prepares it for verification against a Canadian case law database. It keeps a human being in the loop. Rather than letting AI check AI, the tool ensures that a real person makes the final call on whether each case exists.

A verification tool alone is not enough. Courts and tribunals can take three proactive steps to protect self-represented litigants before fictitious citations ever reach a filing:

  • Educate litigants about the specific risk. Say it plainly, in intake materials, on court websites, and at the start of proceedings: AI tools fabricate cases. They invent case names, citations, and holdings that sound real but are not.
  • Point people to real legal research. A self-represented litigant may find ChatGPT long before they find CanLII. Courts should actively direct people to free, reliable legal research tools.
  • Explain what total reliance on AI actually costs. AI does not only fabricate citations. It fabricates entire arguments that sound persuasive and are wrong. The consequence is not just a sanction. It is losing a case whose arguments never had a legal foundation.

Common questions

How many fictitious citations have been found in Canadian courts? As of March 10, 2026, Canadian courts and tribunals have flagged at least 211 non-existent cases cited as real law across 111 decisions and 42 courts and tribunals. The figure is conservative, reflecting only cases that judges caught and wrote about in published decisions.

What are AI-hallucinated citations? They are case references generated by AI tools that look real but point to cases that do not exist. They typically carry a plausible case name, citation, year, and court, which makes them hard to spot without verification.

Do AI tools create fictitious legal cases? Yes. Generative AI tools can fabricate case names, citations, and holdings that appear authentic. In 82 of the 111 decisions in our dataset, the court found or presumed that AI tools produced the fictitious citations.

How can I verify whether a Canadian legal citation is real? CaseCheck lets you upload your case list and prepare each citation for verification against a Canadian case law database. You can also search for cases directly on CanLII, Canada’s free legal research database.

The problem is growing faster than the system can catch it.

Methodology & Sources

Courtready manually searched decisions published on the Canadian Legal Information Institute (CanLII) from January 1, 2024, to March 10, 2026, using keyword searches designed to capture judicial language indicating that a cited authority could not be verified, as well as decisions that discuss the use of artificial intelligence in legal proceedings. The findings are a conservative estimate: decisions not published on CanLII, or that do not use language the search captured, are not reflected. In 82 of the 111 decisions, courts found or presumed AI involvement; in the remaining 29, the source of the non-existent cases was not conclusively established. The full dataset is bilingual, public, and updated weekly at courtready.ca/fictitious-citations-in-canadian-courts.