Hro Banner
May 18, 2026

Practical Tips to Minimize Discovery Risks When Using AI Tools

A growing number of our clients are using AI technology in connection with their work.  At the same time, recent case law has shown that there are scenarios where AI prompts and AI‑generated content are discoverable and are not considered to be privileged communications or protected attorney work product – even when the tools are used to ask legal questions.

First, in U.S. v. Heppner (S.D.N.Y. February 17, 2026), the defendant, who was acting on his own initiative and not at the instruction of his counsel, used an AI tool (Claude) to research legal questions, synthesize factual information relevant to his case, and organize his defensive strategy.  Some of the information provided to the AI tool reflected privileged legal advice that he had received from counsel.  When the FBI later seized the devices containing the prompts and responses, the defendant argued that those documents were privileged.  The court rejected that argument for three reasons: (i) the defendant had used a public version of the AI tool, so there was no expectation that the information provided would remain confidential; (ii) the AI tool is not a qualified attorney; and (iii) the documents were not prepared by or at the behest of counsel.

Second, in Fortis Advisors, LLC v. Krafton, Inc. (Del. Ch. Mar. 16, 2026), the court determined that a buyer had breached an acquisition agreement by wrongfully terminating key employees and seizing control of a subsidiary to avoid paying an earn-out.  In the opinion, the court quoted at length from conversations between the buyer’s CEO and an AI chatbot (ChatGPT) used to develop the plan that was later implemented by the CEO.  The court viewed the chatbot’s recommendations as a window into the buyer’s strategic intent and expressly relied on those exchanges to establish the pretextual nature of the terminations.  The court also found that the CEO had deleted specific, relevant logs from the AI platform.  The ruling highlighted that communications with AI chatbots are easily obtained during discovery, are subject to preservation obligations and can be highly probative for courts evaluating whether a party is operating in good faith.

The following practical tips can help manage these risks going forward:

  • Enterprise-grade, secure AI tools with “closed” systems that prohibit retention and disclosure of user data, such as those in use at Willkie, should always be used. Sharing information that is otherwise privileged or confidential with a public AI platform undermines or waives those protections, particularly where inputs may be used to train the underlying model or may be made available to third parties.
  • Remember that no matter how sophisticated or personalized an AI tool’s responses may feel, the interaction is not a privileged communication because AI is not an attorney.
  • Attorney work product protections may apply if counsel directs the use of AI or where counsel helps to construct the prompts, because it would then reflect counsel’s mental impressions and strategy. However, simply sending counsel outputs from an AI tool after the fact does not make such usage of AI privileged.  Keep a record of the legal purpose and supervising attorney for tasks involving AI, because an audit trail is important if work product protections are challenged.
  • AI-generated output must be reviewed and, where appropriate, revised by the supervising attorney. No AI output should be relied upon without review and sign-off by an attorney.
  • Using AI to transcribe, record, or summarize meetings expands the universe of potentially discoverable materials, and any discrepancies between AI-generated materials and formal minutes will be scrutinized. Be thoughtful about whether such use is appropriate.
  • Consider whether existing litigation hold and preservation protocols adequately account for AI-generated materials, including prompts, outputs, meeting summaries, meeting transcriptions, and related metadata.
  • Establish clear internal rules governing AI use and make them part of regular training.

 

We can work with you to tailor this guidance to your organization.