In the past two months, we have written a few blogs dealing with the use of fake cases by lawyers. Notwithstanding that it is widely known that generative artificial intelligence is hallucinating cases, lawyers apparently continue to rely on this technology to bolster their legal arguments. It is, of course, wrong to mislead the court. The use of fake cases is, in my view, a serious form of misleading a court.
In Hussein v. The Minister of Immigration, Refugees and Citizenship, 2025 FC 1060 (CanLII), the court has ordered that costs will either be awarded against the Applicant or his lawyer because of the use of fake cases.
In this case, the Applicant brought two motions for the admission of new evidence on a judicial review and for the extension of time to file his Application Record.
The Applicant's proceedings before the Refugee Protection Division (RPD) and Refugee Appeal Division (RAD) had been unsuccessful, and accordingly he was seeking to quash the RAD's decision and have the matter remitted for reconsideration.
To support his judicial review, the Applicant wanted to submit new evidence that consisted of certified document issued by the Asyut II District Court in Egypt confirming that he was under investigation for alleged ties to the Muslim Brotherhood. The Muslim Brotherhood was an organization designated as a terrorist group by the Egyptian government.
The Applicant contended that this document was relevant and highly material to the judicial review, and that, due to no fault of his own, he had been unable to obtain the document before the RPD and RAD hearings.
The Applicant deposed in an affidavit that he faced significant legal and bureaucratic barriers in obtaining the document, had no safe way of obtaining proof while inside Egypt, and was only able to obtain the document after the dismissal of his appeal.
The Applicant argued that the new evidence should be admitted because i) it was not available at the time of the RPD and RAD hearings; ii) it was credible and relevant; and iii) it could have affected the outcome if it had been available.
As well, the Applicant argued that the RAD's refusal to consider the new evidence was unreasonable and that the refusal to admit it resulted in procedural unfairness. The Applicant relied on three cases to support his arguments.
In contrast, the Minister stated that the evidence on a judicial review was generally the evidentiary record that was before the decision-maker, subject to three exceptions for the admission of new evidence on a judicial review. However, none of the exceptions was met.
The Minister further submitted that the proper authority for the admission of new evidence was Association of Universities and Colleges of Canada v. Canadian Copyright Licensing Agency (Access Copyright), 2012 FCA 22 ["Access Copyright"], not the authority relied upon by the Applicant.
The Applicant disagreed that Access Copyright was the governing authority and asserted that the more flexible approach cited in its authorities should be adopted.
The court rejected the Applicant's position on the grounds that it was well-understood that a party cannot submit new evidence on an application for judicial review. The court also accepted that there were three exceptions to the general rule, none of which applied to the Applicant's new evidence.
The court explained that it struggled with the authorities relied upon by the Applicant.
One case did not articulate the test presented by the Applicant and did not consider the issue of admissibility of new evidence for the first time on judicial review. The court also found that the Applicant's reliance on other cases was misplaced.
The court was troubled by the Applicant's use of undeclared generative artificial intelligence on the motions. The court noted that in reviewing the Applicant's materials, a number of authorities could not be located and that the court had been required to issue four directions in connection with the need for the Applicant to provide a complete Book of Authorities.
The Applicant failed to comply with the directions. It was only after the fourth direction that the Applicant's lawyer admitted to having relied on Visto.ai (a generative AI tool for immigration practitioners). The lawyer stated that the authorities and citations had not been independently verified as they were understood to reflect well established and widely accepted principles of law. The Applicant's lawyer also argued that the undeclared and unverified AI generated cases had no impact as the Applicant's substantive legal argument was unaffected.
The court was unpersuaded by this argument. The use of AI had to be declared and its output had to be verified by a human.
Although costs in the context of immigration proceedings were not ordinarily awarded, the court concluded that in the circumstances there were special reasons warranting an award of costs, which either might be imposed against the Applicant or the Applicant's lawyer.
The key takeaway is that the use of generative AI for legal research is risky. Although the use of AI is unlikely to be avoided, the potential for hallucination is, as described in this case, "well known". Accordingly, litigants must ensure that the Practice Directions of courts are followed in regard to the use of AI and lawyers should ensure that any cases retrieved from an AI tool are verified. Failure by a lawyer to declare that they have used AI or verify citations is a serious issue and will expose a lawyer to a potential personal costs award in addition to other possible sanctions. A PDF version is available to download here.
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.
