Using artificial intelligence to write an asylum claim is already prohibited in Canada.
Since Monday, September 7, a directive from the Canadian refugee court has been in effect: no one may use artificial intelligence to generate or modify the narrative supporting a person’s claim. The rule makes no distinction between those with legal representation and those representing themselves, and those who violate it risk having their documents rejected or their credibility questioned by the court itself.
What exactly does the new rule prohibit?
The directive, signed on July 10 by Manon Brassard, chair of the Immigration and Refugee Board of Canada (IRB, the administrative tribunal that resolves refugee claims), is direct: “The parties must not use AI to generate or materially change content that reports or reflects evidence of a person.”
In practice, that covers the most important part of the case file: the statement of claim, known as the Basis of Claim, affidavits, witness testimony, and also any photos and videos submitted as evidence. None of that can be generated by an AI tool or rewritten by one.
What is allowed are the usual minor functions: spell checker, grammar checker and document formatting, as long as they do not change the content.
What if the AI was used for something else?
This must be stated in writing and within the document. When AI generated or materially modified any text, the file must include a statement with a precise wording: that artificial intelligence was used, that all generated content was reviewed and verified, and the name of the person who verified it.
The use of AI for transcription or translation must always be declared, without exception, and the human translator remains obligated to sign their own declaration, as before. This point deserves special attention for the Hispanic community: translating a birth certificate, police report, or threatening letter with an automated tool and submitting it without notification is already an offense under the new rule.
What happens if someone doesn’t comply?
The list of consequences reserved by the IRB is extensive, and none are minor. It may refuse to accept the document, draw negative conclusions about the credibility of the person submitting it, report the case to the professional body that regulates the lawyer or consultant involved, restrict that representative’s ability to appear, and take any other action it deems necessary.
The second of these consequences is the most serious for the applicant. In an asylum case, credibility is almost everything: if the court begins to doubt that the story told is genuine, the entire case is compromised, even if the persecution that motivated it is real.
Where did this concern come from?
Back in April, federal officials had already warned that AI was being used to fabricate content in immigration and asylum applications. An IRCC spokesperson acknowledged at the time that cases had been detected in which artificial intelligence helped generate fraudulent applications, with references to nonexistent court rulings, legal precedents that did not support the claims, and stories of persecution invented from beginning to end.
The side effect is that these cases lengthen the review of all others. Both IRCC and IRB are now using machine learning tools to detect false narratives, manipulated documents, and altered photographs, although neither uses artificial intelligence to make the final decision on a case: the oral hearing remains the primary filter.
What should change from today on the kitchen table
Many Latin American applicants arrive at this stage without a lawyer, with limited English or French, and with the perfectly understandable temptation to ask a free tool to organize their story or translate their documents. The guideline explicitly applies to these individuals, not just to professionals.
The practical interpretation is simple: the story must be written in the words of the person who lived it, even if they are clumsy or contain errors. For document translation, a human translator who signs off on the translation is preferable, and if AI was involved for any reason, declaring it from the outset is much cheaper than explaining it later in court.
Editorial Staff Mauricio Navas Talero LJI Reporter
