Lead Lawyer Responsible For Use Of AI By Third Party Contractor (Reddy v. Saroya)

Prior to joining Gardiner Roberts LLP in 1999, I was an outsource lawyer, providing legal support services to other lawyers throughout Ontario. Instead of dealing directly with a lawyer's client, I would only report to the lawyer who had engaged me. Any legal document that I produced for court would be filed in the name of the lawyer for whom I worked. In this regard, the lawyer who engaged me was ultimately responsible for my work product and any errors that might be contained therein.

The responsibility of a lawyer for the work of a third party contractor or outsource lawyer was recently commented upon in Reddy v. Saroya, 2025 ABCA 322.

In this case, the Court of Appeal of Alberta was required to determine if an order declaring the appellant, S, in civil contempt for failing to provide adequate responses to several undertakings should be reversed. To support the appeal, S's lawyer filed a factum, which had been produced by a third party contractor. The factum contained a written argument and cited case law.

While the Court ultimately found that the chambers judge had erred in his application of the test for civil contempt and that the appeal should be allowed in part, the Court expressed significant concerns with the original factum that had been filed in support of the appeal because it contained references to fabricated cases.

Lawyers for the respondent had identified the use of potentially fabricated cases because S's original factum failed to include hyperlinks or copies of the cases that were relied upon. The respondent explained in its original responding factum that seven of the cases cited in S's factum could not be found, including six that were purportedly decided by the Court of Appeal. The respondent identified that the non-existent cases raised a potential artificial intelligence issue.

When confronted by the respondent's lawyers, S's lead lawyer was adamant that the cases existed, but that the citations were wrong. This proved to be incorrect.

Eventually, the lead lawyer sought permission to file an amended factum. He explained that he had subcontracted the drafting of the original factum to a third party, that it had been delivered to him late, and that he had been unable to properly verify the authorities used.

Before the appellate court, the lead lawyer further explained that at the time the original factum was due it was the holiday season, he was ill, and he was very busy.

Accordingly, in the circumstances, the lead lawyer submitted that he should not be required to pay any costs or penalty for the time spent by the respondent's lawyers to search for the non-existent cases and to respond to the original factum. The lead lawyer further submitted that he and his client could work out the details of paying any costs award imposed, such that an award against him was unnecessary.

The Court disagreed with the lead lawyer and stated that in the circumstances it was considering imposing a costs award against him, up to the amount requested by the respondent.

In justifying this decision, the Court noted that rule 3.1-2 of the Law Society of Alberta's Code of Conduct required lawyers to perform all legal services to the standard of a competent lawyer.

Commentary under this rule noted that lawyers were required to "develop an understanding of, and ability to use, technology relevant to the nature and area of a lawyer's practice and responsibilities."

The Court also noted that the Law Society of Alberta had published "The Generative AI Playbook" and had emphasized that lawyers who used generative AI were required to understand their potential benefits and risks.

Lastly, the Court stated that the Alberta Courts had issued a Notice to the Public and Legal Profession entitled "Ensuring the Integrity of Court Submissions When Using Large Language Models" to reinforce the integrity and credibility of legal proceedings.

This Notice urged users to exercise caution when using AI tools and to rely exclusively on official court websites, commonly referenced commercial publishers, or CanLII when referring to cases, statutes or commentary. Moreover, this Notice clearly required human involvement, stating "[i]n the interest of maintaining the highest standards of accuracy and authenticity, any AI-generated submission must be verified with meaningful human control."

None of the precautions seem to have been taken by the lead lawyer or the third party contractor who had assured the lead lawyer that an AIlarge language model had not been used in preparing the original factum.

This case is instructive for multiple reasons. The decision confirms that lawyers are responsible for the citation of fake cases in court filings, and that lawyers who hire third party contractors are responsible for the work product of a third party contractor.

As well, the decision shows that lawyers cannot rely on a myriad of excuses to avoid potential costs sanctions for the use of fake cases in court filings.

In my view, the most significant takeaway from this case is that the use of generative AI for legal research and preparing factums (and notices of motion and the like) should be strongly discouraged within the legal profession, if not banned in its entirety.

Although some might disagree with my view, there is simply no need to utilize generative AI to conduct legal research, particularly when CanLII is available to the entire profession for free and commercial publishers have developed AI search capabilities within their own defined sand box of verifiable sources. As is being seen more and more often, the use of generative AI and large language models outside of these legitimate sources is a very bad short-cut for finding answers to legal issues and for spending the time to produce a well-written and persuasive factum. 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.