Canada’s Federal Court is the only Canadian court to have put a formal definition of the word in writing. Its notice on the use of artificial intelligence in court proceedings defines it in a footnote: “‘Hallucination’ is a term used to refer to facts, citations, and other content generated by AI that are not true, and have been fabricated by AI in response to a prompt or request.”
Ontario’s Consolidated Civil Provincial Practice Direction describes the same problem from the other side of the same event, in almost the same words: “Hallucinations can consist of non-existent cases, mischaracterizations of case law, and fabricated quotations.” That direction does not stop at what counsel files. It goes on to flag a parallel duty for expert witnesses: “counsel and litigants should also be aware of subrules 03 (2.1) and (2.1.1) regarding the obligation of experts to verify the authenticity of the authorities or other documents or records referred to in their report(s).” A hallucinated citation is therefore not only a lawyer’s problem — an expert’s own report carries an equivalent authenticity duty.
Outside the courtroom, Canada’s Cyber Centre frames the same failure as an ongoing operating risk rather than a one-time mistake to catch and move past, listing “model drift, hallucinations, bias and overreliance” together among the things a deployed AI system has to be continuously watched for.
A paralegal asks a drafting tool to find supporting case law for a motion, and the tool returns a citation with a real-looking case name, a court, and a paragraph number — none of which appears in any reporter, because the tool composed a plausible-sounding one rather than looking one up. Under the Federal Court’s rule above, whoever files the resulting document must still declare that AI content was used; under Ontario’s rule, the lawyer signing the factum is personally certifying they are satisfied every cited authority is genuine. Neither rule is satisfied by asking the same tool to double-check its own citation. It is satisfied by opening the case and reading it.
See also large language model and retrieval-augmented generation for what raises and lowers this risk.
This is one term in a plain-English glossary on how AI actually works and where it fits in a Canadian business.
