There is no special Canadian rule for AI-generated evidence. What exists is a general framework for electronic documents, written long before generative AI, and a set of very recent court notices about how AI may be used to prepare the materials that go in front of a judge in the first place.
Key takeaways
Start with the general rule, because it existed long before anyone asked whether AI changes anything. Section 31.1 of the Canada Evidence Act puts the burden squarely on whoever wants an electronic document admitted as evidence: that person must prove the document is authentic. The rule does not distinguish between a document a person typed and one AI generated or assembled — authenticity has to be shown either way, by evidence capable of supporting the finding that the document is what it claims to be.
Section 31.2 satisfies the best-evidence rule for an electronic document in one of two ways: proof of the integrity of the system that recorded or stored it, or an applicable evidentiary presumption. Section 31.3 then supplies that presumption — in the absence of contrary evidence, an electronic documents system is presumed to have been operating properly, and therefore to have preserved the document’s integrity, unless there are reasonable grounds to doubt it. Nothing in either provision turns on whether AI was involved in creating the content; both turn on the system that recorded and stored it.
None of the above is about AI appearing as evidence in a case — it is about AI having been used to help write the factum, the affidavit, or the legal argument submitted to the court. The Federal Court’s Notice on the Use of Artificial Intelligence in Court Proceedings requires a Declaration whenever content submitted for litigation was created or generated by AI, in the first paragraph of the document, unless AI was used only to suggest changes or critique material a human then implemented. Notably, the notice carves out expert reports specifically: AI use there is disclosed instead in the methodology summary required under the Expert Witnesses Code of Conduct.
The same Federal Court notice defines two terms worth knowing precisely, in its own footnotes, because they describe two different ways AI can taint what ends up in front of a court. A “hallucination” is “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” — a defect in AI-assisted legal argument.
A “deepfake” is “AI-generated images of human subjects that either replace one person’s likeness convincingly with that of another, or that do not exist in real life” — a defect that can attach to the underlying evidence itself, not just the argument built around it. The Court’s stated response to both is the same principle: a “Caution” principle, urging counsel to rely only on well-recognized, reliable sources.
The Consolidated Civil Provincial Practice Direction carries a provision that gets less attention than the general AI-hallucination warning surrounding it: alongside the certification a lawyer signs confirming the authenticity of every authority cited in a factum, Ontario’s civil procedure rules separately place an obligation on experts to verify the authenticity of the authorities, documents or records they rely on in their own reports. That is a rule built for exactly the kind of exposure an AI-assisted expert report can create — a citation or a supporting record that turns out not to be genuine — even though the rule predates generative AI entirely.
A party wants to submit a screenshot of a business record, generated and formatted by an internal reporting tool with an AI-driven summarization feature, as evidence in a civil claim. Under the framework above, the question is not whether AI touched the file — it is whether the party can show the electronic documents system that produced and stored it was operating properly, satisfying Section 31.2’s best-evidence rule through Section 31.3’s integrity presumption, and separately whether the underlying document is authentic under Section 31.1. If the same party’s lawyer also used a generative tool to draft the factum introducing that evidence, that is a second and entirely separate disclosure question, governed by whichever court’s AI notice applies, not by the Evidence Act provisions above.
It is easy to blur AI produced this evidence with AI helped write the document about this evidence, because both can be true in the same case and both involve the word AI. But they are answered by different bodies of rules, aimed at different risks: the Evidence Act provisions protect against a court relying on a digital record that cannot be trusted to be what it claims to be, while the court AI notices protect against a lawyer or litigant submitting argument built on fabricated legal authority. A litigant who satisfies one has not automatically satisfied the other, and a lawyer preparing a file that touches both should treat them as two separate checklists rather than one combined AI question.
Separate again from both of the above is a question the Canadian Judicial Council addresses directly: not what a litigant submits, but what a judge does. Its own guidelines open on a single non-negotiable line: “It is fundamental to the independence, impartiality and integrity of the judiciary for a judge to exercise the powers of office without undue or unauthorized reliance upon non-judges”, and state plainly that no judge may delegate decision-making authority “whether to a law clerk, administrative assistant, or computer program, regardless of their capabilities” (Canadian Judicial Council, Guidelines for the Use of Artificial Intelligence in Canadian Courts, September 2024). treadstonelaw.ca’s explainer on electronic evidence in Ontario courts covers the general authentication question from the litigant side.
No. There is no rule that excludes evidence simply because AI was involved in creating or processing it. It is tested the same way any electronic document is tested — authenticity and the best-evidence rule — under the Canada Evidence Act's general framework.
Increasingly yes, in courts that have adopted a specific notice — the Federal Court's Declaration requirement is the clearest example — but this is a disclosure rule about how the litigation materials were prepared, separate from whether any AI-touched evidence itself is admissible.
A hallucination is fabricated content, typically a false citation or fact, generated by AI and embedded in legal argument. A deepfake is a fabricated or altered depiction of a real person, which can show up as the disputed evidence itself rather than as an error in how a lawyer's submissions were drafted.
Related: how AI changes accounting and audit work and does AI training infringe copyright.
Authenticating what a system produced, and where its inputs came from, is exactly the kind of question a diligence process is built to answer.
