As businesses increasingly integrate generative artificial intelligence tools such as ChatGPT, Claude, Microsoft Copilot, and Gemini into their daily operations, those tools are creating a new—and potentially revealing—source of electronically stored information. Employees are using AI to draft documents, analyze contracts, summarize meetings, evaluate business decisions, calculate damages, research competitors, and sometimes simply think through difficult problems.
Those interactions may not disappear when litigation begins.
In fact, courts are already confronting disputes over whether AI prompts, responses, chat histories, and related materials are discoverable. The emerging decisions suggest that traditional principles of relevance, privilege, work product, and preservation will generally apply. But they also demonstrate why businesses and their litigation counsel should begin thinking about AI conversations in much the same way they think about emails and text messages.
One reason AI history may become particularly valuable in litigation is that a prompt can reveal information that does not appear in the final work product.
Consider a hypothetical contract dispute.
A company executive asks ChatGPT:
“We promised the customer delivery by September 1, but we knew in July that our supplier could not meet the deadline. Help me draft an email explaining the delay without admitting that we knew about it earlier.”
The resulting email may be carefully written and relatively innocuous.
The prompt is another matter entirely.
It potentially reveals knowledge, timing, intent, and an effort to frame subsequent communications. Depending upon the claims and defenses at issue, that information could be far more probative than the final email.
The same concept can arise in many types of commercial disputes. An employee might ask an AI tool whether taking certain customer information to a new employer violates a confidentiality agreement. A corporate officer might ask AI to analyze whether a contemplated transaction breaches an existing contract. Someone preparing a damages claim might ask an AI platform to develop different assumptions until it produces a desired number.
Recent commentary on AI discovery has therefore emphasized that litigators may need to look beyond the document ultimately created and consider the AI interactions that preceded it.
There is not yet a comprehensive body of appellate law governing AI discovery. But several recent federal decisions illustrate how courts are applying familiar discovery doctrines to this new technology.
In United States v. Heppner, 820 F. Supp. 3d 292 (S.D.N.Y. 2026), a criminal defendant used Anthropic’s Claude platform to research potential charges and defenses in connection with a pending criminal investigation. Some of those exchanges incorporated information relating to conversations the defendant later had with counsel.
The court held that the defendant’s communications with Claude were not protected by the attorney-client privilege or the work-product doctrine. The communications were made to a publicly available AI platform—not to counsel—and were disclosed to the third-party provider under circumstances that defeated the requisite confidentiality. Nor did the defendant’s subsequent decision to share the materials with counsel retroactively confer privilege on communications that were unprotected when created.
Although Heppner arose in a criminal prosecution, its reasoning has broader implications. A businessperson who uses a public AI platform to analyze the facts of an emerging commercial dispute should not assume that those exchanges will receive the same protection as confidential communications with counsel.
On the other hand, in Warner v. Gilbarco, Inc., 820 F. Supp. 3d 629 (E.D.Mich. 2026), the court held that a pro se plaintiff’s use of ChatGPT to assist with litigation did not waive work-product protection. The court reasoned that AI tools, such as ChatGPT, are not “persons” under the legal framework governing work-product waiver. The Warner court reasoned that waiver of work-product protection requires disclosure to an adversary or in a manner likely to reach an adversary, which did not occur in this instance.
The court determined that the materials generated using ChatGPT reflected litigation-related mental impressions and strategies, qualifying them for work-product protection under Federal Rule of Civil Procedure 26(b)(3). The court emphasized that the use of AI tools in litigation does not inherently waive such protection, as disclosure to an AI tool does not equate to disclosure to an adversary.
The court’s decision was grounded in the principle that work-product protection is not waived unless materials are disclosed in a manner that makes them accessible to an adversary. The court did not suggest that all uses of AI tools in litigation would automatically qualify for work-product protection; rather, the protection applied in this case because the materials were prepared in anticipation of litigation and were not disclosed to an adversary.
The two cases are not necessarily inconsistent. But rather illustrate an important point: the discoverability of an AI conversation will depend heavily upon who created it, why it was created, what it contains, and what privilege or work-product protection applies.
So, what does the evolving landscape mean practically?
Perhaps the most immediate issue for businesses is not admissibility but preservation.
Once litigation is reasonably anticipated, parties generally must take reasonable steps to preserve potentially relevant electronically stored information. Historically, litigation holds have focused on familiar sources such as email, documents, text messages, mobile devices, shared drives, and collaboration platforms.
AI systems now raise an additional question: Should relevant AI conversations also be preserved?
In appropriate cases, the answer may be yes.
If employees used generative AI to discuss the transaction, contract, employee, competitor, product, or events at the center of a dispute, those conversations may contain potentially relevant information.
That does not mean every AI interaction throughout an organization must automatically be preserved. Discovery remains subject to relevance, proportionality, privilege, accessibility, and other traditional limitations.
But companies and their counsel should at least determine whether relevant custodians used AI tools in connection with the subject matter of the dispute.
That inquiry may need to happen early. Some AI platforms or corporate configurations may retain conversation histories differently from traditional email systems, and retention settings can vary. Recent guidance concerning AI discovery has therefore recommended that litigation-hold procedures specifically consider AI prompts, outputs, and interaction logs where appropriate.
The emerging AI discovery cases should not cause companies to prohibit useful technology reflexively. Generative AI can provide substantial efficiencies.
But businesses should recognize that AI interactions are not necessarily a private scratchpad existing outside the normal rules of litigation.
Companies should consider incorporating several issues into their AI governance and litigation-readiness programs:
Twenty years ago, many companies underestimated how dramatically email would change litigation. Later came text messages, social media, Slack, Teams, and other forms of electronic communication.
Generative AI appears likely to be the next evolution.
The law surrounding AI discovery is still developing, and courts will continue sorting through important questions involving relevance, privilege, work product, proportionality, confidentiality, and preservation.
But one practical rule is already emerging.
Before typing something into an AI platform, businesses and their employees should consider whether they would be comfortable seeing the same words projected onto a screen in a courtroom.
Because someday, they might be.
Disclaimer: This article is for informational purposes only and does not constitute legal advice. The application of discovery, privilege, and preservation rules depends upon the particular facts and circumstances of each matter.
To read more from Tim, visit Tim Horton’s LinkedIn.
