Canada's Immigration and Refugee Board has brought into force a practice notice prohibiting parties from using AI to generate or materially change basis of claim narratives, affidavits and witness statements before any of its four divisions

The prohibition, and how far it reaches

Section 3 is the operative restriction, and the notice puts it plainly: "Parties must not use AI to generate or materially change content that reports or reflects a person's evidence - such as basis of claim narratives, affidavits, or witness statements. These must be based on a person's own knowledge and experience."

The notice then gives its own examples of prohibited use: drafting a basis of claim narrative, affidavit or witness statement with an AI tool; rewriting a person's story; and creating or altering photographs, videos, screenshots or other corroborative evidence. That last limb is the one to read twice. It puts image and video evidence inside the same prohibition as the written narrative.

There is a stated carve out, and it is narrow. The prohibition does not prevent minor assistive functions such as spellcheck, grammar correction or formatting assistance, provided those tools do not generate the evidence or make material changes to it. The notice offers no definition of materially change, so the boundary between grammar correction and rewriting a story is left to be argued case by case.

Definition matters here. A note to the notice defines AI broadly, as systems or technologies that can simulate human-like capabilities such as learning, reasoning, problem-solving and administrative or managerial decision-making to assist or automate processes, and says this includes generative AI. Parties, per another note, include counsel and self-represented parties.

The declaration, and who has to be named in it

Section 4 applies to everything else. Where a party uses AI to generate or materially change content for a document submitted to the IRB, the notice sets out the wording to include: "Artificial intelligence (AI) was used to create or materially change text in this document. All content generated by AI, and the authenticity of any cited case law or other legal authorities, has been reviewed and verified by [name]."

Two features of that text carry the weight. It requires a name, not a firm or a file number, and it puts the authenticity of cited authority inside the same sentence as the AI-generated content. A person signs for both.

Parties may add detail about which software was used, for what purpose, and whether AI was used for the whole document or only particular paragraphs. The notice frames that as optional. The minor assistive functions carve out repeats here: no declaration is needed for spellcheck, grammar correction or formatting assistance, provided those tools do not generate or materially change the wording.

Transcription, translation, and the human translator

Section 4 also carries a separate and stricter rule for two functions. Use of AI for transcription or translation is to be disclosed always, without the materiality qualifier that governs the general declaration.

The notice reminds parties that the Rules in all divisions require translated documents to be accompanied by a declaration from a human translator, and adds that the translator is to declare whenever AI was used to assist in translation. That places a disclosure obligation on a person who may not be a party or represented.

Elsewhere the notice sets expectations rather than prohibitions. Parties are told to be ready to answer questions about their AI use in each proceeding and, when requested, to be able to establish the authenticity and accuracy of their documents. They are told not to pad submissions with AI-generated material unrelated to the facts of the case, with the notice citing a 2025 Federal Court decision, and to weigh the privacy risk of putting sensitive information into tools that may lack sufficient security features.

What happens if a party ignores it

Section 5 lists five consequences, and every one of them is framed as something the IRB may do rather than a fixed penalty. The Board may refuse to accept a document for filing or otherwise decline to rely on it, and may draw negative inferences about the credibility of the evidence.

The third is the one practitioners should note. The IRB may disclose information to counsel's professional regulatory authority, under its existing policy on disclosing information regarding the conduct of authorized representatives to regulatory bodies. The route to a law society already existed; this notice identifies AI-related non-compliance as a possible basis for using it.

The remaining two are the IRB restricting or prohibiting counsel from appearing before it, and taking any other necessary action. The notice does not rank these, set thresholds, or say which breach attracts which response.

On verification, the notice ties the general duty to sources. Parties using AI to build a case are told to ensure the content is accurate, verifiable and trustworthy, including systematic verification of legal principles and case citations using reliable sources. A note defines reliable sources as legal databases, official government websites, commonly referenced commercial publishers or trusted public services such as CanLII, and expressly excludes AI-generated summaries or other AI tools found inside those sources.

Two dates, and why we print both

The notice was signed on 10 July 2026 and comes into effect on 7 September 2026. Both statements appear in section 6 of the instrument itself. The in-window date is the instrument's own stated coming-into-force date, and we report it the way a commencement provision is reported, rather than presenting the signature as recent.

A third date sits in the page furniture. The published page carried a date modified of 8 September 2026 when we read it, and its embedded issued field read 8 July 2026, which matches neither the signature date nor the effective date. Page metadata is not the instrument. We take section 6 as authoritative and record the discrepancy rather than quietly resolving it.

One sourcing note belongs here. The French language twin of this notice is not published at the mirrored French path. That address returns HTTP 404 and a stub page that redirects to the IRB's own page-not-found screen. The French text is published under a different French slug, which the English page's language toggle carries, and we opened it there. It runs the same six numbered sections in the same order under the same headings. We quote only the English, and we do not report the French wording of any provision.

What we did not verify

What we opened: the English practice notice in full on the IRB site, including sections 1 to 6 and all four explanatory notes, the prescribed declaration wording, the consequences list and the effective date and signature block. We also opened the French twin at its own French slug and confirmed it carries the same six sections.

What we did not open: the IRB Policy on Disclosing Information Regarding the Conduct of Authorized Representatives to Regulatory Bodies that section 5 routes to; the division Rules that require a human translator's declaration; and the 2025 Federal Court decision cited in the notice's own footnote. We describe each only as this notice describes it.

One typography note, because it touches a quotation. The published notice sets off the examples in the section 3 sentence with a typographic dash, and we reproduce it as an ASCII hyphen, as we do with the apostrophes elsewhere. That is a house typesetting convention and not a change to any word.

What we refuse to claim: we do not say a breach leads to a law society referral, because the notice says the IRB may disclose information to a regulator and sets no trigger, threshold or sequence. We do not say what counts as materially changing a document, because the notice does not define it. We do not say this is the first Canadian tribunal practice notice of its kind, because the document makes no such claim and we did not test it. We do not state how the notice interacts with any division's existing rules on document filing, because we did not read them.

Informational analysis for working professionals, not legal advice. Confirm how any rule applies to your situation with qualified counsel.

Related briefings

• Canadian law societies on lawyers' AI duties
• Ontario's LSTH costs award over AI hallucinated citations
• Canadian sanctions for AI hallucinations in filings
• North America AI regulation tracker
• AI Regulation News hub