It is no surprise that many organizations are keen to implement artificial intelligence (or more commonly, “AI”) throughout their workflows. AI has the potential to dramatically speed up banal, day-to-day tasks and synthesize large volumes of complex information. However, recent case law from the United States and the United Kingdom casts a shadow over this broad utility. Early cases show that AI-generated discussions and prompts can be producible in litigation and may be used to prove a party’s actions or intentions. For organizations that have embraced AI tools, these developments warrant careful attention.
Consider first the recent decision in Fortis Advisors, LLC v. Krafton, Inc. In that case, the Delaware Court of Chancery was faced with allegations the defendant, Krafton, had sought to avoid certain contractual earnout obligations. AI chat logs in evidence showed that Krafton's CEO had used AI to devise — and subsequently implement — a strategy to seize operational control of a subsidiary and avoid the earnout payment. The Court effectively treated the CEO’s lengthy AI chat logs as evidence of the defendant’s strategic intent — and bad faith. This decision demonstrates that discussions with an AI chatbot should be treated with the same care as any other communications, and underscores the evidentiary risks of using AI tools for strategic advice.
Similar trends are emerging in the United Kingdom. In Anthony Malcolm Cork & Anor v. Mark Smith, the High Court of Justice was faced with a letter from a law firm seeking relief on the basis of a statutory provision that did not exist as quoted. On discovering this, the Court asked for witness statements from the law firm. The firm produced the relevant lawyer’s AI chat transcripts, which showed that the chatbot had first quoted the (hallucinated) statutory provision, but had also warned the lawyer to “verify the wording” by consulting the statute directly. The Court found that the lawyer’s discussions with the AI chatbot evidenced their lack of care and judgment, including by failing to heed the chatbot’s warnings to check the statutory provision directly. The Court did praise the law firm for its candour, noting that the firm had been “full and frank in providing the transcripts of the AI chat.” In the circumstances, the Court declined to initiate contempt proceedings and instead found that referral to the regulator was the proportionate response for counsel involved.
Finally, in Conservation Law Foundation, Inc. v. Shell Oil Co. (decision currently under challenge), the U.S. District Court for the District of Connecticut ordered disclosure of all AI prompts used by the plaintiff’s expert witness in preparing her report. The expert alleged that she had merely used AI to cull the list of documents produced by the defendant in the litigation, so that she could limit her review to only the most relevant documents. The Court held, however, that the process of selecting documents via AI forms part of an expert’s methodology, and is therefore within the scope of discovery. This decision, if built upon by other courts, could be used to find that even largely administrative uses of AI may be fair game for discovery.
Taken together, these three cases suggest that chat logs with AI tools are increasingly probative in litigation. Courts are willing to examine AI interactions to understand how an expert reached certain conclusions, to illuminate a party’s intentions or strategy, or to assess the actions of a litigant or their counsel. In short, AI chat transcripts are becoming yet another tool in the court’s toolkit to uncover facts, strategic motivations and methodologies, and parties should treat them accordingly.
Although there have not yet been any reported cases of this nature in Canada, the prospect of similar rulings is very real. Under Ontario’s Rules of Civil Procedure, parties to a civil action must disclose “[e]very document relevant to any matter in issue in an action that is or has been in the possession, control or power of a party to the action.” Critically, “document” is defined broadly for the purposes of discovery; it includes “a sound recording, videotape, film, photograph, chart, graph, map, plan, survey, book of account, and data and information in electronic form.” Given this expansive definition, it is no great leap to conclude that relevant conversations with a chatbot could fall within its scope.
Ontario is currently proposing changes to the Rules of Civil Procedure to deal with AI evidence, including a proposal that will require a party who seeks to adduce evidence containing AI to disclose that fact, and will allow the court to require that party to disclose information (including inputs) used to produce the AI evidence. This may include prompts used by experts relying on AI to form their opinions. The cases above from the United States and the United Kingdom show us that some courts are already doing this, and are otherwise already willing to rely on AI prompts as evidence. The prudent approach would be to consider AI-generated documents and AI prompts and conversations as producible. In other words, don’t communicate with your favourite AI chatbot in a way you wouldn’t want before a judge.
Of course, AI chats and prompts may not be producible where litigants have a valid claim of privilege over the documents. For our insights on how privilege interacts with AI, see our recent bulletin, AI and Legal Privilege: Practical Considerations From Emerging Case Law.
For more information, please contact the authors or any other member of our Litigation & Dispute Resolution group.
