Our research on fictitious case law appears at footnote 39 of the Government of Canada’s new AI transparency discussion paper. Submissions close September 23, 2026.
On July 23, 2026, Innovation, Science and Economic Development Canada (ISED) opened a public consultation on AI transparency. The discussion paper that anchors it, Enhancing trust in artificial intelligence through increased transparency, cites our study.
The citation sits in one of five case studies in the paper. The other four cover deepfake fraud, a dealership chatbot that made a car buy-back offer the dealership then tried to revoke, a prompt injection vulnerability in a widely deployed enterprise product, and the risks of agentic commerce. Ours is the one on legal documents.
What the Discussion Paper Says
The paper reports that 44 Canadian courts and tribunals issued 132 decisions between January 2024 and March 2026 in which a party relied on a case that does not exist. In 96 of those decisions, the adjudicator found or inferred that AI had produced the fictitious case.
Those numbers come from our CanLII study.
The point the paper draws from them is that AI output can read as confident and authoritative while being wrong. It notes that the risk grows when a system is used for a task it was not built for, or when its limitations are not well understood.
Where the Numbers Stand Now
The study period closed on March 26, 2026. The count has moved since.
The current count is also a floor. The database captures only those cases where an adjudicator caught a fictitious citation and said so in a published decision. Those that pass the bench, opposing counsel, and the parties themselves are invisible to this kind of research.
What the Consultation Is Asking
The paper does not propose rules. It sits under AI for All, the national AI strategy released in June 2026, and its purpose is to work out whether transparency measures are needed, and if so, whether they should arrive through legislation, standards, voluntary codes, or procurement requirements. The paper is explicit that the strategy commits the government to the work but does not settle the outcome. That is the stage at which outside input has the most effect.
The government has identified five areas where it is considering action:
- Detecting and identifying AI-generated content
- Knowing when you are interacting with an AI system
- Information about AI systems
- Tracking serious AI incidents
- Tracking the activity of AI agents
Two bear directly on the justice system. The third asks what information users and businesses need in order to understand what an AI system can and cannot reliably do. The fourth asks how a serious AI incident should be defined, who should collect that information, and what should be done with it.
The paper closes with four cross-cutting questions, one of which invites readers to name areas the five categories do not cover.
How to Participate
There are two ways to respond. The anonymous survey is on the ISED website. Written comments go by email to AIConsultations-ConsultationsIA [at] ised-isde.gc.ca.
Why It Is Worth Filing
Lawyers, adjudicators, court staff, legal clinics, and self-represented litigants have direct experience of what happens when fabricated authority reaches a courtroom. That experience will not be in the record unless someone puts it there.
We will be making a submission. If you work in or around the justice system, we would encourage you to make one too.