Yes — but what you must declare, and to whom, changes from one Canadian court to the next. The duty to verify never changes.
Key takeaways
Yes. No Canadian court has banned counsel from using AI to prepare filings. What has changed is not permission but accountability: several courts now require you to say that you used it, and every court that has spoken assumes you personally verified everything it produced. The rules are not uniform, so the operative question in any given matter is which court you are in.
Two duties recur. The first is disclosure — telling the court, and the other side, that AI generated content in a document filed for the purposes of litigation. The second is verification — checking every authority against a reliable source before the document leaves your office. Disclosure varies by forum. Verification does not vary at all.
The Federal Court’s notice and interim principles on artificial intelligence ask parties to inform the Court, and each other, when documents prepared for the purposes of litigation contain content created or generated by AI. The declaration goes in the first paragraph of the document — the first paragraph of a memorandum of fact and law, for instance — and states whether AI was used throughout or only in identified paragraphs.
The Court draws a line that matters in daily practice. A declaration is required where the role AI played resembles that of a co-author. It is not required where AI was used merely to suggest changes, provide recommendations, or critique content a human had already written and then considered and implemented manually. Editing help is not co-authorship.
The notice does not reach certified tribunal records, and expert reports are handled separately through the disclosure of methodology expected under the expert witnesses code of conduct. Where a file arrives from previous counsel, new counsel are expected to make best efforts to establish whether earlier material contains AI-generated content, and to declare it where they have reason to believe it does.
The same page carries the Court’s interim principles on its own use of AI, which state that the Court will not use AI — and specifically automated decision-making tools — in making its judgments and orders without first holding public consultations.
Ontario has taken a different route: less about declaring, more about guaranteeing. The Consolidated Civil Provincial Practice Direction devotes a section to the use of AI for court proceedings, and states that it is the responsibility of all counsel and litigants to guarantee accuracy when preparing materials, particularly when using AI, regardless of whether they directly interacted with the technology. Counsel and litigants must exercise careful, informed and ongoing oversight, and the practice direction says plainly that the court will not tolerate inadvertence.
It is explicit about the failure mode. Misuse most often occurs when counsel or litigants carelessly rely on fictitious authorities generated by AI — hallucinations, which the practice direction describes as non-existent cases, mischaracterisations of case law and fabricated quotations.
It then connects that to the rules of procedure. A factum must include a statement, signed by the party’s lawyer or by someone the lawyer has specifically authorised, certifying that the person signing is satisfied as to the authenticity of every authority cited in it. Authorities published on a government website or by a government printer, on CanLII, on a court’s website, or by a commercial publisher of court decisions are presumed authentic absent evidence to the contrary. That signed certification is the point where a fabricated citation is either caught or is not.
The consequences are written down rather than implied. The court’s powers include public reprimand of counsel or the litigant, cost orders, adjourning a hearing or dismissing the matter, the initiation of contempt proceedings, and, for counsel, referral to the Law Society of Ontario.
The Court of King’s Bench of Manitoba issued a short practice direction in June 2023 on the use of artificial intelligence in court submissions. It requires that where AI has been used in preparing materials filed with the court, the materials indicate how it was used. Same instinct, different mechanism: Manitoba asks how, the Federal Court asks for a declaration in a set form and place, and Ontario asks you to guarantee the result.
The practical consequence for a firm appearing in more than one forum is that an AI policy cannot be a single sentence. It has to be a table: for every court and tribunal you appear before, what must be declared, in what words, and where in the document it goes.
Court directions sit on top of professional-conduct obligations, not instead of them. The Law Society of Ontario publishes guidance for licensees on generative AI — a white paper, a practice note on professional obligations, and quick-start checklists — through its technology resources for lawyers, and Ontario’s civil practice direction points counsel back to both the Rules of Professional Conduct and that white paper.
Conduct rules differ by province, so a firm licensed in more than one has more than one rulebook to reconcile. None of them makes AI the author of a filing. It drafts, summarises and suggests; a licensed lawyer reads, verifies and signs, and remains answerable for every word.
A three-lawyer litigation boutique wants the drafting speed without the exposure. The workflow that survives the practice direction looks like this.
One. The model drafts structure and prose from the firm’s own memo and the record — never from its recollection of the law. Nothing it produces about a case is treated as a source.
Two. Every proposition of law is matched to an authority a person retrieved from a source the rules presume authentic. If a proposition cannot be matched, the proposition changes; the citation is never made to fit.
Three. Pin cites and paragraph references are checked against the judgment itself, because a real case with a fabricated paragraph reference fails the same certification as an invented case.
Four. The certifying statement is signed by the person who did the checking, not by whoever is closest to the printer.
Five. If AI generated content rather than critiquing it, the forum’s declaration requirement is applied in the form that forum specifies.
The time saved sits almost entirely in step one. Steps two to four are where the work moved. They did not disappear, and a firm that budgets as though they did will eventually explain itself to a judge.
Do I have to declare AI use in every Canadian court?
No, and that is the trap. Requirements differ by forum: the Federal Court wants a declaration in the first paragraph, Manitoba’s Court of King’s Bench wants the materials to indicate how AI was used, and Ontario’s Superior Court concentrates on accuracy and the signed authenticity certification. Check the specific court before filing.
Does using AI only for research trigger the Federal Court declaration?
The declaration is aimed at content created or generated by AI in the filed document, and the Court says it is not required where AI merely suggested changes or critiqued human-written content. Research assistance you then verify and write up yourself is a different act from AI-authored paragraphs. The verification duty is unaffected either way.
What happens if a fabricated authority reaches the court?
Ontario’s practice direction sets out the range: public reprimand, costs, adjournment or dismissal, contempt proceedings, and referral to the Law Society of Ontario. The response depends on the facts, but the list is not theoretical.
A 30-minute call is enough to tell you whether AI pays for itself here.
