Author:
Artificial intelligence (“AI”) tools are now part of everyday life. Many people use AI platforms like ChatGPT and other generative AI tools to answer questions about a variety of topics, including legal issues. Individuals and businesses may be using AI to research their own cases before speaking with a lawyer, or even after they have retained a lawyer, to verify or expand upon the legal advice received.
While this may seem convenient, using AI for legal research or advice can create risks that both clients and lawyers should be aware of, including:
- Hallucinating fake cases, facts, and other information
- Waiving legal privilege over information shared with AI
AI Can Invent Legal Authorities
One of the biggest risks with generative AI is something called “hallucination”. This happens when the AI system produces information that is fabricated or inaccurate. In a legal context, that may include citing cases that do not exist. While some may assume they can avoid this problem by simply searching for the case to confirm that it exists, that does not eliminate the risk in its entirety. AI tools can also cite real cases with correct citations but generate quotations that do not actually exist in the decision at all.
Even in situations where AI cites real cases and quotations, it can still misinterpret the decision, misstate the case’s rationale or legal principles, or overstate the case’s relevance. This means that lawyers and clients alike should not rely on AI alone for legal research or legal advice without verification.
In Kapahi Real Estate Inc. v. Elite Real Estate Club of Toronto Inc., 2026 ONSC 1438, counsel submitted a factum that cited real cases with correct neutral citations that could be located on CanLII, an online Canadian case law database. However, the factum also included seven fake quotations from several different court decisions that were not actually found in those decisions.
The court noted that the most obvious explanation for the fake quotations was that AI had been used to generate the factum. However, the court did not make a formal finding that AI was used, as the court had not heard a full argument on the issue. Nevertheless, the court referred the decision to the Law Society of Ontario to investigate the wrongdoings of the lawyer. The court also referred to a previous case, Ko v. Li, 2025 ONSC 6785, in which the court considered finding a lawyer in contempt of court for citing fake AI-hallucinated cases.
In short, both lawyers and laypersons must be aware that including false or misleading AI-generated cases or other legal information in either written or oral submissions to the court may result in professional discipline, imposition of fines for the misuse of court time, and even criminal or regulatory proceedings.
Using AI Could Result in a Waiver of Solicitor-Client Privilege
Secondly, using AI for legal research could result in a waiver of solicitor-client privilege. A recent U.S. court decision highlights the risk of using AI to obtain legal information and advice.
In United States v. Heppner, 1:25-cr-00503(JSR), 2026 WL 436479 (S.D.N.Y. Feb. 17, 2026), the Southern District of New York District Court held that certain written exchanges the defendant had with a generative AI platform were not protected by attorney-client privilege. In this case, the defendant, Heppner, was charged with, amongst other things, securities fraud, wire fraud, and falsifying corporate records, arising from his alleged misconduct as an executive of several corporate entities, including a publicly traded company. The allegations included that Heppner had defrauded investors out of more than $150 million.
The Federal Bureau of Investigation executed a search warrant at Heppner’s home and seized numerous documents and electronic devices. Among the seized materials were approximately 31 documents that memorialize communications that Heppner had with the AI platform Claude. Heppner had used Claude to prepare reports that outlined his defence strategy and what he might argue in anticipation of a potential conviction.
Heppner’s legal counsel asserted privilege over these AI-generated documents, arguing that Heppner had inputted into Claude information that Heppner had learned from counsel, that he had created these AI documents for the purpose of speaking with counsel to obtain legal advice, and that he had subsequently shared these AI documents with counsel.
The court noted that it is well established law in the U.S. that attorney-client privilege attaches to, and protects from the disclosure, communications (1) between a client and his or her attorney (2) that are intended to be, and in fact were, kept confidential (3) for the purposes of obtaining or providing legal advice.
However, in this case, the AI documents were not communications between Heppner and his counsel. Because Claudeis not an attorney, that alone disposes of Heppner’s claim of privilege. Second, the communications memorialized in the AI documents were not confidential. The AI platform’s written privacy policy provided that the data of both the user’s inputs and the platform’s outputs were collected and used to train Claude, and that the platform reserved the right to disclose such data to a host of third parties, including governmental regulatory authorities. By entering the information into a third-party platform, Heppner voluntarily disclosed it to a company that could store, process, or potentially share that data. Third, Claudedisclaims providing legal advice and, thus, the documents were not created for the purpose of obtaining legal advice from counsel. It is important to note that in this case, Heppner’s attorneys had not directed or recommended to Heppner that he use AI for the purpose of speaking to his legal counsel or otherwise obtaining legal advice.
The court also emphasized that non-privileged communications do not become privileged because they are later shared with counsel.
Although Heppner is a U.S. case, solicitor-client privilege in Canada is premised on similar principles. In Canada, the general rule is that voluntary disclosure of a privileged document to a third party normally constitutes a waiver of privilege. Accordingly, clients should avoid uploading confidential information, such as e-mails or documents sent by their legal counsel, to AI systems to prevent any potential waiver of solicitor-client privilege. Lawyers should also warn their clients of these risks.
To summarize, once a client voluntarily shares confidential information with a third party, including an AI platform, the law may treat the communication as no longer confidential, which can result in a waiver of solicitor-client privilege. Most generative AI tools state in their terms of service that they store user prompts and the system-generated responses. That means a client who copies their lawyer’s advice into an AI tool could inadvertently expose that communication to discovery by the other party.
For assistance with your legal needs, please do not hesitate to reach out to any member of the LK Law Team.
This article is intended to be an overview of the law and is for informational purposes only. Readers are cautioned that this article does not constitute legal or professional advice and should not be relied on as such. Rather, readers should obtain specific legal advice in relation to the issues they are facing.
