Skip to content
Disputes & Class Actions

AI in Litigation: Practice Directions, Hallucinations and Evidence

AI in litigation is reshaping how Canadian courts receive submissions and evidence. This guide covers the Federal Court's AI declaration notice, Ontario's rule requiring lawyers to certify that authorities are authentic, costs consequences for hallucinated cases, the discoverability of prompts and chatbot logs, deepfake evidence and the risks AI poses to arbitral awards.

Lawyer verifying citations as part of AI in litigation safeguards in a Canadian court filing

Key takeaways

  • The Federal Court requires a declaration where filed content was directly provided by AI.
  • Ontario factums must certify that every cited authority is authentic and quotations accurate.
  • Courts have ordered costs against lawyers personally for hallucinated authorities.
  • Prompts, outputs and AI meeting notes can be discoverable documents.
  • Electronic evidence must be authenticated; suspected deepfakes need early forensic work.

AI in litigation: where Canadian courts stand

AI in litigation is no longer hypothetical. Lawyers and self-represented parties use generative tools to research, draft and summarize; companies use AI to review documents in discovery; and evidence itself increasingly includes AI-generated or AI-altered material. Canadian courts have responded with practice directions requiring disclosure of AI use, new rules requiring lawyers to certify that cited authorities are real, and costs consequences where fabricated cases reach the court. For businesses involved in disputes, the practical questions are what their counsel must disclose, how AI-related records may be discoverable, and how to deal with evidence that might be synthetic.

Lawyer verifying citations as part of AI in litigation safeguards in a Canadian court filing
AI in litigation brings disclosure duties, verification rules and new evidentiary questions for courts and parties.

The rules differ by court, so the starting point in any proceeding is the current practice direction of the court or tribunal concerned. The examples below focus on the Federal Court and Ontario, which have the most developed written rules.

Practice directions on AI-generated submissions

The Federal Court's notice to the parties and the profession, updated on 7 May 2024, requires a declaration where a document prepared for litigation and filed with the Court contains content directly provided by AI, or where AI's role resembles that of a co-author. The declaration goes in the first paragraph and identifies the paragraphs concerned.

The notice exempts uses such as AI that only suggests edits or critiques human-written text, routine automation such as voice recognition, and certified tribunal records; expert reports should disclose AI use in their methodology instead. The Court states that a declaration on its own will not attract an adverse inference, and that the party signing a document remains responsible for its accuracy. The Court has also published interim principles on its own use of AI (last updated 29 September 2025), including a commitment not to use AI or automated decision-making tools to make judgments or orders without first holding public consultations.

Ontario has gone further by amending its Rules of Civil Procedure. Under r. 4.06.1(2.1), a factum must include a statement signed by the party's lawyer certifying that the signer is satisfied that every authority cited is authentic and, following a 2026 amendment (O. Reg. 275/26), that every quotation from an authority accurately reproduces its text.

Self-represented parties must sign the statement themselves (r. 4.06.1(2.1.1)). Authorities published on government websites, CanLII, court websites or by commercial publishers of court decisions are presumed authentic absent evidence to the contrary (r. 4.06.1(2.2)). Expert reports carry a parallel certification about the authenticity of authorities and documents the expert relies on.

ForumMain rule on AI in litigationWhat filers must do
Federal CourtNotice on the use of AI in court proceedings (updated 7 May 2024)Declare AI-generated content in the first paragraph; verify content; rely on recognized sources
Ontario Superior Court and Court of AppealRules of Civil Procedure, r. 4.06.1(2.1)–(2.2)Lawyer certifies every cited authority is authentic and every quotation is accurate
Ontario expert evidenceExpert report certification of authenticity of authorities and documents relied onExperts must check what they cite, including AI-sourced material
Other courts and tribunalsPractice directions vary by province and tribunalCheck the current direction before filing; some require disclosure, others caution only

Hallucinated authorities and costs sanctions

Generative tools can produce case names, citations and quotations that look real but do not exist. Courts in several Canadian jurisdictions have dealt with filings that cited such "hallucinated" authorities, and the consequences have included costs ordered against the lawyer personally, adjournments, referrals to the law society and, in serious cases, contempt-style show-cause proceedings. Courts have been clear that the duty to verify sits with the lawyer who signs the document; reliance on an AI tool is an explanation, not an excuse. Self-represented litigants have also had costs awarded against them for wasting court and opposing-party time on fictitious citations.

For companies, the risk is reputational and financial as much as procedural. A client whose counsel cites fake authorities may lose credibility with the court on the merits, face adverse costs and see its case delayed. Companies should ask their litigation counsel directly how AI is used on their files and what verification steps are in place, and should apply the same expectation to in-house teams drafting affidavits or regulatory submissions.

Practical point

Adopt a simple rule for any document going to a court, tribunal or regulator: every authority and quotation is checked against an official or recognized source by a person, and the check is recorded. That habit satisfies Ontario's certification rule, the Federal Court's verification expectation and most other directions.

Discoverability of prompts and chatbot logs

Records generated by AI tools are "documents" for discovery purposes if they are relevant. Prompts, outputs, chat histories and AI-generated meeting summaries can all be producible, just like emails. Several scenarios are now common in commercial disputes:

  • Business decisions informed by AI. If a pricing, credit or hiring decision was made with an AI tool, the prompts and outputs may be relevant to whether the decision was reasonable or discriminatory.
  • AI meeting notes. Automatic transcription and summary tools create records of conversations that participants may have assumed were informal.
  • Customer-facing chatbots. Chat logs may be central evidence in consumer and misrepresentation claims; see our insight on chatbot liability.
  • Counsel's own AI use. Prompts and outputs created by or for counsel to prepare litigation may attract litigation privilege, but material created in the ordinary course of business generally will not.

The practical consequence is that AI tools must be covered by litigation holds and retention policies. Our insight on document preservation and litigation holds explains how, and our guide to privilege covers the risk of placing privileged material in public AI tools.

AI-generated evidence and deepfakes

Courts must also decide what to do with evidence that is, or is alleged to be, generated or altered by AI: images, audio recordings, videos and documents. Existing evidence law provides the framework.

Under s. 31.1 of the Canada Evidence Act, a person seeking to admit an electronic document must prove its authenticity by evidence capable of supporting a finding that it is what it purports to be; s. 31.2 addresses the best evidence rule through proof of the integrity of the electronic documents system. Provincial evidence statutes contain similar rules, and courts can also draw on the general rules for real evidence, which require proof that a photograph or recording fairly and accurately represents what it purports to show.

Authentication thresholds are low, which is why deepfakes are a concern. A party who suspects fabrication should raise the issue early, preserve the original files and metadata, demand production of native files, and retain a forensic expert. Courts can weigh the reliability of evidence even after admitting it, and expert evidence on manipulation detection is increasingly common. Parties relying on recordings or images should be ready to prove their chain of custody from the device that captured them.

AI in arbitration and the risk to awards

Arbitration raises its own issues. Arbitrators are expected to decide disputes personally, and parties may challenge an award if a tribunal delegated its decision-making function, including to an AI tool, or if a party was denied a fair opportunity to respond to material the tribunal relied on. Under the UNCITRAL Model Law framework that applies to international arbitration in Canada, procedural unfairness and decisions outside the scope of the submission are recognized grounds for setting aside an award. Parties can reduce the risk by agreeing at the outset, in procedural orders, on whether and how AI may be used by counsel, experts and the tribunal, and on disclosure of that use. Our international arbitration practice addresses those protocols.

Illustration: a hypothetical supplier in an Ontario commercial dispute receives the other side's factum citing two appellate decisions that cannot be found. Its counsel notifies opposing counsel and the court, which adjourns the motion, orders the factum corrected and awards costs of the wasted hearing against the lawyer who signed the certification. The court also directs both sides to describe their verification process for future filings. The dispute itself is delayed by months, and the client whose lawyer filed the factum starts the rescheduled hearing with its credibility damaged, a cost that no costs order fully repairs.

A governance checklist for AI in litigation

Most of the risks described above can be managed with ordinary governance. Companies that are regularly involved in disputes, and the law departments that manage them, should consider the following steps for AI in litigation:

  1. Know the tools. Maintain a list of approved AI tools for legal work, with their data-retention and confidentiality terms, and prohibit consumer tools for privileged or confidential material.
  2. Set expectations for outside counsel. Engagement letters and litigation guidelines should require counsel to disclose how AI is used on the file, confirm compliance with the relevant court's directions, and verify every authority and quotation.
  3. Extend litigation holds. Hold notices should name AI chat histories, prompt libraries, transcription services and AI-generated summaries as categories to preserve.
  4. Agree discovery protocols. Where technology-assisted review or generative AI will be used to review documents, the parties can agree on validation methods, sampling and disclosure of the approach, which reduces later disputes about completeness.
  5. Plan for synthetic evidence. Identify forensic experts in advance, preserve native files and metadata for key recordings and images, and document chain of custody from the outset.
  6. Train witnesses and staff. Employees should understand that AI outputs used in business decisions may be produced and examined in court, and should not use AI tools to draft their own witness statements or affidavits without counsel's involvement.

Budgets and timelines also change. Responsible use of AI in litigation can reduce the cost of document review and first drafts, but verification, disclosure and expert work add steps that should be planned rather than discovered at the filing deadline. Clients are entitled to ask how AI-driven efficiencies are reflected in fees, and to expect that savings never come at the cost of accuracy before the court.

Regulators and law societies are also issuing guidance on professional responsibility and AI, generally emphasizing competence, confidentiality, supervision and candour with the court. Those obligations apply to the lawyers on a file, but companies benefit when their counsel's approach to AI in litigation is documented and consistent across matters.

How TRW & Co helps

We advise companies on AI in litigation as part of our commercial litigation and artificial intelligence work. That includes AI-use protocols for litigation teams and experts, verification workflows that meet court certification rules, litigation holds that capture AI records, discovery strategy for prompts and chatbot logs, and responses to suspected synthetic evidence with forensic experts. Court proceedings in Canada require Canadian-licensed counsel, and we work alongside them. This page is general information rather than legal advice. To discuss an AI-related litigation issue, contact us.

Frequently asked questions

Do Canadian courts require disclosure of AI use in filings?

Some do. The Federal Court requires a declaration in the first paragraph of a document where content was directly provided by AI. Ontario requires lawyers to certify in factums that every cited authority is authentic and every quotation accurate. Other courts and tribunals have their own directions, so the rule depends on the forum.

What happens if a lawyer cites a case invented by AI?

Canadian courts have responded with costs orders against the lawyer personally, adjournments, corrective filings and referrals to the law society. The lawyer who signs a document is responsible for its accuracy, and reliance on an AI tool does not excuse a failure to verify authorities.

Are AI prompts and chat logs discoverable?

If they are relevant and not privileged, generally yes. Prompts, outputs, chat histories and AI meeting summaries are documents like emails. Material prepared by or for counsel for litigation may be protected by litigation privilege, but ordinary business use of AI usually is not.

How do courts deal with suspected deepfake evidence?

The party relying on electronic evidence must prove it is what it purports to be, for example under s. 31.1 of the Canada Evidence Act. A party alleging fabrication should preserve originals and metadata, demand native files and retain a forensic expert. Courts can also give less weight to evidence whose reliability is doubtful.

Can an arbitral award be challenged because AI was used?

Potentially, if the use meant the tribunal did not decide the dispute itself or a party was denied a fair opportunity to address material relied on. Parties should agree on AI-use rules in early procedural orders to reduce the risk of later challenges.

Sources

  1. Federal Court: Notice to the parties and the profession – use of AI in court proceedings (7 May 2024)
  2. Federal Court: Artificial intelligence
  3. Rules of Civil Procedure, R.R.O. 1990, Reg. 194, r. 4.06.1
  4. Canada Evidence Act, R.S.C. 1985, c. C-5, ss. 31.1–31.2

General information only, not legal advice. References to statutes, courts and authorities describe published law and practice at the date shown.