AI in litigation: AI prompts and outputs not categorically protected
Norton Rose Fulbright has been closely monitoring a growing body of 2026 federal and state court decisions addressing whether and when generative artificial intelligence (GenAI) prompts and outputs may be entitled to privilege protection. Our initial article discussed US v. Heppner (Heppner),1 a federal decision that found the GenAI prompts and outputs at issue were not privileged. Our follow-up article compared Heppner to two other federal cases that reached different conclusions, Warner v. Gilbarco, Inc. (Warner)2 and Archie Morgan v. V2X (Morgan).3 This post discusses another recent federal decision on the issue.
On October 21, 2026, the US District Court for the Southern District of Florida issued a second order after discovery hearing in Fontainebleau Florida Hotel, LLC v. Botach4 (Fontainebleau), addressing, among other things, whether a defendant’s prompts and outputs (referred to as “communications” in the order) with a third-party GenAI platform were protected by the work product doctrine. The court held that the doctrine does not categorically protect those communications and rejected the defendant’s blanket objection to producing them. However, in doing so, the court also clarified that using a third-party platform does not automatically waive work product protection either.
This article (1) summarizes the Fontainebleau order, (2) provides background on the discovery dispute that led to the order, (3) highlights the court’s work product analysis and (4) identifies potential implications for litigants and lawyers.
Summary of Fontainebleau order
- The court found that neither Heppner nor Warner establishes a categorical rule on whether work product protection applies to third-party GenAI communications–instead, neither discoverability nor protection should be presumed, and litigants must still look to the “ordinary requirements” of Rule 26(b) and the work product doctrine.
- A party’s independently generated communications with a third-party GenAI platform are not protected simply because the party is a litigant or because the resulting content may bear on the litigation.
- Work product protection turns on whether a particular GenAI communication was prepared because of litigation rather than, for example, to create public-facing statements. The party asserting protection must support it on a document-specific basis, including through a privilege log.
The Fontainebleau lawsuit and the discovery dispute
Fontainebleau Florida Hotel, LLC sued Shmuley Boteach Jacob (a/k/a Shmuel Botach) for allegedly defamatory statements about the hotel. By the time of the order, discovery had closed, but the district court had reopened it on a limited basis, authorizing “discovery limited to damages.”5
The hotel served Jacob with a request for production and four proposed subpoenas to AI service providers, each seeking his communications with third-party GenAI platforms since December 3, 2024, including communications concerning the hotel, the litigation, the hotel’s counsel, employees or representatives, court orders entered in the case and the presiding judges.
Jacob asserted a blanket objection to that discovery, arguing, among other things, that the requests sought materials protected by the work product doctrine such as GenAI-assisted litigation preparation, legal research, draft analyses and communications concerning defense strategy. He relied principally on Warner in arguing that litigation-related materials do not automatically lose work product protection merely because a party used a third-party GenAI platform to prepare them.
Work product: The doctrine does not categorically protect AI communications
After determining that some of the hotel’s requests could be relevant, the court explained that Federal Rule of Civil Procedure 26(b)(3)(A) protects relevant documents and tangible things prepared in anticipation of litigation or for trial by or for a party or its representative, and that this protection is not limited to materials prepared by counsel.
The court agreed with Heppner that independently generated GenAI communications do not become work product merely because the user is represented by counsel. But it declined to read Heppner as holding that everything a represented party prepares on its own is categorically unprotected, because Rule 26(b)(3)(A) extends to qualifying materials prepared “by or for” a party (emphasis added).
The court then followed Warner in recognizing that GenAI-assisted litigation preparation may qualify for work product protection and that using a third-party AI tool does not necessarily waive that protection. But it also decided that Warner did not support Jacob’s blanket objection. Such materials remain protected only if they were “created principally or exclusively to assist in anticipated or ongoing litigation,”6 and a document’s creation after litigation commenced does not, standing alone, show that it was prepared for the litigation. The party asserting protection bears the burden of establishing that each of the particular materials qualifies, using evidence specific to those materials rather than a generalized assertion that they concern litigation.
Applying these principles, the Court concluded that GenAI communications Jacob generated to create, revise or disseminate public-facing content about the hotel, its personnel or its counsel, rather than to prepare his defense, would not satisfy Rule 26(b)(3)(A).
The court noted, however, that Jacob may still assert protection over any particular communication that was prepared because of the litigation and otherwise satisfied Rule 26(b)(3), provided he identifies the withheld material and describes it sufficiently for the hotel to assess the claim, as Federal Rule of Civil Procedure 26(b)(5)(A) requires, including through a privilege log.
Implications of Fontainebleau order
Prior to Fontainebleau, recent decisions on this topic–including Heppner and Warner–largely focused on the public nature of certain GenAI tools that permit third parties to train on prompts and outputs and whether the use of these public tools vitiates confidentiality and thus automatically waives the attorney-client privilege. Fontainebleau does not raise this issue, relying on both Heppner and Warner to support the finding that GenAI prompts and outputs are neither automatically discoverable nor automatically protected. Instead of adopting a categorical rule, it suggests that courts and parties should apply the “ordinary requirements” of Federal Rule 26(b) to determine whether there is a sufficient basis for each GenAI interaction to support work product protection.
Taken together, these decisions support the principle that materials created with GenAI tools, even publicly available ones, may be entitled to work product protection, but only on a proper showing. Going forward, litigants should be aware of three implications from the Fontainebleau order.
First, work product protection for GenAI communications does not solely turn on the tool used. Courts will likely examine the purpose of each use, so litigants who use GenAI tools for purposes that mix litigation preparation with other activities–such as drafting public statements, responding to media inquiries or pursuing personal projects–should anticipate that courts will parse those communications individually rather than treat an entire category of documents as protected.
Second, a party asserting work product protection over GenAI communications bears an evidentiary burden. As with any other material, a generalized assertion that communications “concern litigation” or involve “AI-assisted litigation preparation” will likely not suffice; the party must be prepared to identify the withheld material and substantiate, on a document-specific basis and through a privilege log, that each communication was created because of the litigation.
Third, Fontainebleau illustrates that the content and purpose of GenAI prompts and outputs, not merely their existence, can determine whether they must be produced. Litigants and their counsel should consider how a party’s use of GenAI tools, including for communications unrelated to privileged legal strategy, may later become discoverable if that use is relevant to the claims and defenses at issue.
We will continue to monitor developments as courts and rulemakers provide further guidance. For questions regarding GenAI governance or litigation matters, please contact us.
Special thanks to law clerk Emma Santizo for assisting in the preparation of this article.