Florida Magistrate Narrows AI Chat Discovery in Defamation Suit

Florida Magistrate Judge Narrows AI Chat Discovery in a Defamation Case

A federal magistrate judge in Miami ruled on October 21, 2026 that a defamation defendant's chatbot conversations are neither automatically discoverable nor automatically protected. The 25-page order narrowed the plaintiff's requests, required a privilege log, and allowed limited subpoenas to four AI providers.

Information last verified on October 23, 2026. This is a developing story; we update it as the record changes.

Status: This is an interim discovery ruling by a United States magistrate judge, entered October 21, 2026, not a merits decision and not binding precedent. A follow-up discovery hearing is tentatively set for October 26, 2026 at 1 p.m. and may be canceled if the parties resolve their disputes. Damages discovery closes November 9, 2026. The defendant's interlocutory appeal of a separate September 18, 2026 order modifying the restrictions on his extrajudicial speech is pending before the Eleventh Circuit, and on October 22, 2026 the district judge denied a related reconsideration motion without prejudice for lack of jurisdiction while that appeal is pending. Nothing in the October 21 order resolves what the four AI providers themselves may argue.

Jurisdiction scope: This article covers one federal discovery order from the United States District Court for the Southern District of Florida and the Federal Rules of Civil Procedure and Florida damages statutes it applies. It does not address state privacy statutes, chatbot terms of service, or law-enforcement access to AI records. For the substantive rules behind the underlying claim, see our guide to defamation law in Florida.

What Happened

United States Magistrate Judge Marty Fulgueira Elfenbein entered a 25-page Second Order After Discovery Hearing, ECF No. 289, on October 21, 2026 in Fontainebleau Florida Hotel, LLC v. Botach, No. 1:25-cv-20251 (S.D. Fla.), captioned Case No. 25-CV-20251-KUNTZ/Elfenbein. The plaintiff hotel had served a seventh request for production and circulated four proposed subpoenas to AI service providers, seeking defendant Jacob Shmuel Botach's communications with AI platforms since December 3, 2024 across several subject categories. The court overruled his blanket objections to three categories and to part of a seventh, sustained them as to three others and the rest of the seventh, denied the plaintiff's request to force him to consent to the subpoenas, and permitted narrowed Rule 45 subpoenas to the providers. Anything withheld as work product must now appear on a privilege log. The ruling is an interim discovery decision, not a merits ruling, and it is not binding on any other court.

The order follows a discovery hearing held October 8, 2026. A First Order After Discovery Hearing, ECF No. 281, issued October 9 and resolved most of the disputes, but took two under advisement: the defendant's blanket objection to the seventh request for production and his objections to the four proposed AI-provider subpoenas. The First Order noted that neither side had offered any legal authority at the hearing, so the court ordered expedited supplemental briefing limited to ten pages per side. The parties filed their memoranda on October 12, 2026 at ECF Nos. 282 and 283.

The seven requested categories were not identical. Six appeared in both the request for production and the subpoenas: AI communications since December 3, 2024 concerning the Fontainebleau Hotel; this litigation; counsel for the plaintiff; any employee, owner or representative of the hotel; any court order entered in the case; and the presiding judicial officers. A seventh category, in the subpoenas only, sought communications in those categories that had been deleted since their creation. The court noted in a footnote that the seventh request's internal cross-reference to "Requests Nos. 1 through 7" appeared to be a typographical error for Requests Nos. 1 through 6.

The push for third-party records had a specific motive. According to the First Order, the plaintiff told the court it wanted the providers' own records "due to concerns about Defendant's possible spoliation of such responsive AI-related materials." That is the practical reason the deleted-communications category exists at all, and it is a familiar pattern in electronic discovery fights over spoliation of evidence.

The procedural posture matters to how the requests were judged. On September 17, 2026, District Judge K. Michael Moore adopted a report and recommendation on the plaintiff's motion for sanctions, struck the defendant's answer and affirmative defenses, dismissed his counterclaim with prejudice, and entered default judgment against him as to liability on the first amended complaint (ECF No. 261). That order reset the damages trial to the two-week period beginning February 22, 2027 and allowed the parties to "conduct discovery limited to damages for an additional thirty (30) days." The case was transferred to District Judge Jeffrey T. Kuntz on September 28, 2026, and Judge Kuntz later extended the damages-discovery deadline through November 9, 2026 without expanding its scope. So the fight over AI chat logs is a fight about the amount of damages, not about whether the defendant is liable.

The headline holding sits at page 21, at the end of a section captioned "The Work-Product Doctrine Does Not Categorically Protect Defendant's AI Communications":

"The Court emphasizes that a party's communications with and use of AI does not automatically establish either discoverability or protection. The analysis turns on the ordinary requirements of Rules 26(b)(1), 26(b)(3), and 26(b)(5). Defendant may not withhold all communications within the permitted categories based on a generalized assertion that they concern litigation, and Plaintiff may not obtain genuinely protected litigation-preparation materials merely by describing them as AI communications."

Second Order After Discovery Hearing, ECF No. 289, at 21.

The conclusion at pages 24 and 25 does five things. It overrules the blanket scope objections to Request Nos. 1, 3, 4 and 7 (in part) to the extent those requests seek nonprivileged AI communications concerning the hotel, the plaintiff's counsel, or any employee, owner or representative of the hotel. It sustains the objections to Request Nos. 2, 5, 6 and 7 (in part), so no production is required on those. It requires compliance with Rule 26(b)(5)(A), including a sufficient privilege log, for anything withheld on work-product grounds. It denies the plaintiff's oral motion to compel the defendant to consent to the subpoenas, ECF No. 288. And it permits Rule 45 subpoenas to the four providers limited to communications since December 3, 2024 about the hotel, the plaintiff's counsel, or any employee, owner or representative of the hotel, after the parties meet and confer on how any responsive materials should be produced. The order adds that "Nothing in this Order compels a nonparty provider to produce information or resolves objections that may be raised by a subpoena recipient."

The order does not name the four AI service providers, and neither does the First Order.

What the Law Actually Says

Discovery scope comes from Federal Rule of Civil Procedure 26(b)(1), which lets parties "obtain discovery regarding any nonprivileged matter that is relevant to any party's claim or defense and proportional to the needs of the case," weighing factors including the amount in controversy, the importance of the discovery, and whether the burden outweighs the likely benefit. The same rule adds that "Information within this scope of discovery need not be admissible in evidence to be discoverable." That sentence is doing real work here: the court repeatedly said it was deciding discoverability, not what a jury will eventually hear.

Work product comes from Rule 26(b)(3)(A), which provides that "Ordinarily, a party may not discover documents and tangible things that are prepared in anticipation of litigation or for trial by or for another party or its representative." The protection traces to Hickman v. Taylor, 329 U.S. 495 (1947), and United States v. Nobles, 422 U.S. 225 (1975), which the order cites at 510-11 and 238-39 respectively for the proposition that the doctrine exists to protect the adversarial process by preserving the privacy a party needs to prepare.

Three features of that rule explain why a chatbot session may or may not qualify.

First, the protection is not limited to lawyers. The text reaches materials prepared "by or for another party or its representative," and the order cites Eleventh Circuit decisions in Tambourine Comercio Internacional SA v. Solowsky, 312 F. App'x 263, 284 (11th Cir. 2009), and Republic of Ecuador v. Hinchee, 741 F.3d 1185, 1192-93 (11th Cir. 2013), for the point that work product is not attorney-exclusive. A self-represented or client-generated document can qualify.

Second, the test is purpose, not timing. The order quotes the standard that litigation "need not necessarily be imminent . . . as long as the primary motivating purpose behind the creation of the document was to aid in possible litigation," and that the material must have been "created principally or exclusively to assist in anticipated or ongoing litigation." A document created after a case was filed does not become work product just because of when it was made, and the party claiming the protection carries the burden of showing it applies.

Third, the protection is qualified, not absolute. Rule 26(b)(3)(A) allows discovery of work product where the requesting party "shows that it has substantial need for the materials to prepare its case and cannot, without undue hardship, obtain their substantial equivalent by other means," while Rule 26(b)(3)(B) still requires the court to protect "the mental impressions, conclusions, opinions, or legal theories of a party's attorney or other representative concerning the litigation."

Put those together and the question for any given chat session is what it was for. The order's own formulation is that communications "independently generated by Defendant to create, revise, or disseminate content concerning the Hotel, its personnel, or Plaintiff's counsel are not transformed into attorney work product merely because Defendant is a litigant." Its reasoning on the other side is equally plain: "The Court must examine the purpose and circumstances of the particular communications rather than presume protection from the medium through which they were created."

The order discusses the two reported 2026 decisions the parties relied on. It describes United States v. Heppner, 820 F. Supp. 3d 292, 298-99 (S.D.N.Y. 2026), as concluding that certain AI communications independently generated by a represented defendant were not work product because they were not prepared at counsel's direction and did not reflect counsel's strategy. The court found Heppner persuasive insofar as it recognizes that independently generated AI communications do not automatically become protected merely because counsel represents the user, while declining to read it as establishing that everything a represented party prepares on its own is categorically unprotected. It describes Warner v. Gilbarco, Inc., 820 F. Supp. 3d 629, 636-37 (E.D. Mich. 2026), as recognizing that AI-assisted litigation preparation may qualify for protection and that using a third-party AI tool does not necessarily waive it, but held that Warner does not support a blanket objection. In the court's words, "Neither decision establishes a categorical rule governing all AI communications."

The mechanics of withholding come from Rule 26(b)(5)(A). A party claiming privilege or trial-preparation protection must "expressly make the claim" and "describe the nature of the documents, communications, or tangible things not produced or disclosed and do so in a manner that, without revealing information itself privileged or protected, will enable other parties to assess the claim." That is the privilege-log requirement the order imposed.

The consent ruling turns on the difference between Rule 34 and Rule 45. Rule 34(a) lets a party request items "in the responding party's possession, custody, or control." The order reasoned that the rule requires a responding party to produce responsive nonprivileged materials within its control but "does not expressly require a party to execute an authorization allowing a third party to disclose records directly to an adversary," and that the plaintiff cited no authority for its request. Analogizing to the split in the case law on compelling medical-record authorizations, and citing Williams v. Carnival Corp., No. 18-21654-CIV, 2020 WL 854809, at *1-2 (S.D. Fla. Feb. 11, 2020), the court declined to compel consent where the information could be sought through ordinary third-party discovery.

Rule 45 governs subpoenas to nonparties. A recipient commanded to produce documents "may serve on the party or attorney designated in the subpoena a written objection," and on a timely motion the court where compliance is required must quash or modify a subpoena that fails to allow reasonable time, exceeds the geographic limits, "requires disclosure of privileged or other protected matter, if no exception or waiver applies," or "subjects a person to undue burden." That is why the order says permitting issuance does not establish enforceability.

On the damages side, the relevance analysis rests on Florida law and on State Farm Mutual Automobile Insurance Co. v. Campbell, 538 U.S. 408 (2003), which identified reprehensibility as the most important consideration in evaluating the reasonableness of a punitive award. Fla. Stat. § 768.72(2) distinguishes intentional misconduct from gross negligence, and § 768.73(1) sets the general limits: an award ordinarily may not exceed the greater of three times compensatory damages or $500,000. The provision the plaintiff invoked is § 768.73(1)(c): "Where the factfinder determines that at the time of injury the defendant had a specific intent to harm the claimant and determines that the defendant's conduct did in fact harm the claimant, there shall be no cap on punitive damages." Our state guide explains how Florida's punitive damages cap and its exceptions fit together.

The order is careful to separate that inquiry from the constitutional one. Citing Harte-Hanks Communications, Inc. v. Connaughton, 491 U.S. 657, 666-68 (1989), St. Amant v. Thompson, 390 U.S. 727, 731 (1968), and Turner v. Wells, 879 F.3d 1254, 1273 (11th Cir. 2018), it notes that constitutional actual malice concerns knowledge of falsity or reckless disregard for the truth, which is a different question from culpability and intent to harm under the punitive-damages framework. Our explainer on the actual malice standard for public figures covers the constitutional half of that distinction. The order also leans on Herbert v. Lando, 441 U.S. 153 (1979), where the Supreme Court declined to recognize a privilege that would have categorically insulated a defamation defendant's editorial thought processes from discovery.

Analysis: Why This Matters

The following is analysis from the Recording Law Editorial Team.

The most useful thing about this order is what it declines to do. It would have been easy to write either of the two headlines people expect from a case like this: that chatbot logs are private litigation material, or that they are fair game. The court wrote neither. It held that the medium is not the test, applied the same Rule 26 analysis it would apply to a notebook or a draft email, and then did the unglamorous work of sorting seven request categories one at a time.

That sorting is where the reasoning shows. The three categories that survived share a feature: each is defined by a person or entity connected to the alleged injury, namely the hotel, its personnel, and opposing counsel. The three that fell share the opposite feature: each is defined by the litigation apparatus rather than by the plaintiff. The request for every AI communication concerning this litigation is, as the order put it, "exceptionally broad." Requests aimed at the defendant's private reactions to court orders and to the judges were rejected not because such material is sacred but because the connection to the hotel's damages was too attenuated. The court's line is worth noting: "The discovery inquiry is not whether Defendant has criticized the judiciary; it is whether the requested information is relevant and proportional to the damages issues remaining in this action."

The work-product holding is narrower than a casual reading suggests, and narrower in a way that cuts both ways. The court expressly refused to adopt the broadest reading of Heppner, because Rule 26(b)(3)(A) protects materials prepared "by or for another party or its representative" and a self-generated document can qualify. It equally refused to let Warner carry a blanket objection. The result is a purpose test applied document by document, with the burden on the person asserting protection. In practice that means a litigant who used a chatbot both to draft public posts and to think through a defense cannot answer a request by gesturing at the whole account. The privilege-log requirement is the enforcement mechanism, and it is the part of this order with the most immediate operational bite.

Two limits deserve emphasis because they are easy to lose. First, the order reaches only the parties. The providers have not appeared, and the court said in terms that nothing in the order compels a nonparty to produce anything or resolves objections a recipient may raise. A served subpoena is the start of a separate Rule 45 process, with its own objection and motion-to-quash track. Second, this is a magistrate judge's discovery ruling in one case. It is reasoned and citation-dense, but it is not appellate authority, and district judges elsewhere are free to analyze the same question differently. The two 2026 decisions the order discusses already point in somewhat different directions, which is itself the signal: this area is being built one discovery order at a time.

There is also a quieter procedural point. The order's concluding "wrinkle" is that if providers produced directly to the plaintiff's counsel, the defendant would lose the chance to review for work product and log it. Rather than resolve that, the court sent the parties to confer on logistics. That instinct, protecting the review step before third-party data moves, is likely to recur in every case where a litigant's records sit on someone else's servers, and it is a more durable contribution than any holding about AI specifically.

What Happens Next

The record sets several dates. The parties were directed to meet and confer on how the providers should be instructed to produce any responsive materials before revised subpoenas issue, and the court reserved October 26, 2026 at 1 p.m. for a continued in-person discovery hearing if the parties cannot resolve the logistics. The First Order provides that the hearing may be canceled if they do. The defendant's continued deposition is set for November 4, 2026, limited to three hours by agreement and confined to damages. Damages discovery closes November 9, 2026. The damages trial is set for the two-week period beginning February 22, 2027, with a final pretrial conference on February 9, 2027 and pretrial motions due December 10, 2026.

Two tracks run alongside the discovery schedule. The defendant filed a Notice of Appeal, ECF No. 286, challenging the September 18, 2026 order modifying the restrictions on his extrajudicial speech; that interlocutory appeal is before the Eleventh Circuit. The magistrate judge explained in a footnote why it did not divest her of jurisdiction over damages discovery, citing Green Leaf Nursery v. E.I. DuPont De Nemours & Co., 341 F.3d 1292, 1309 (11th Cir. 2003), for the rule that an interlocutory appeal removes jurisdiction only over the aspects of the case involved in the appeal. On October 22, 2026, Judge Kuntz applied the same principle in the other direction and denied the defendant's motion for reconsideration of the September 17 and September 18 orders "WITHOUT PREJUDICE for lack of jurisdiction pending the disposition of the Defendant's interlocutory appeal" (ECF No. 291).

If and when the narrowed subpoenas are served, the providers may object or move to quash under Rule 45, which would open a separate dispute this order does not reach. We will update this page as the docket develops. We are not predicting how any of these motions, objections or appeals will be resolved.

How This Affects You

For anyone wondering whether their own chatbot history is private, the honest answer from this order is that it depends on why the conversation happened, and that litigation is the setting where the question actually gets tested. Nothing here suggests a general rule that AI chats are off limits in civil discovery, and nothing here suggests they are automatically available either.

A few general observations follow from the reasoning rather than from the result. Courts applying Rule 26 ask whether material is relevant to a claim or defense and proportional to the case, and a request framed around the people and conduct at issue is far more likely to survive than one framed around everything a party typed into a chatbot. Work-product protection, where it applies, attaches because a document was made principally to prepare for litigation, and the party asserting it has to show that document by document rather than account by account. On the authority this order discusses, using a commercial AI tool does not necessarily waive a protection that would otherwise apply, and it does not create one either.

It is also worth separating the two places the records live. A party's own copies are reached through ordinary party discovery. The provider's copies are a nonparty's records, reached by subpoena, with the provider free to raise its own objections. A court order to a litigant is not an order to the company. If you are facing a real dispute of this kind, that is a question for a lawyer licensed where your case is pending, not something to resolve from a news article. Readers tracking how state law is developing in this space may also want our overview of AI legislation in Florida.

This is general legal information, not legal advice. It covers one federal discovery order from the United States District Court for the Southern District of Florida, the Federal Rules of Civil Procedure it applies, and Florida's punitive damages statutes, and it reflects sources verified on October 23, 2026. Laws change, this story is developing, and the record may change at the October 26, 2026 hearing or afterward. Consult a lawyer licensed in your jurisdiction about your specific situation.

Related articles

• Defamation laws by state
• AI laws and regulation in the United States
• A defamation claim over Google AI Overviews moves forward

---

Last updated: 2026-10-23. This is a developing story; details verified as of 2026-10-23.

Frequently Asked Questions

Did a court rule that AI chat logs are discoverable?

Not as a general proposition. The October 21, 2026 order in Fontainebleau Florida Hotel, LLC v. Botach (S.D. Fla.) says the opposite: "a party's communications with and use of AI does not automatically establish either discoverability or protection" (ECF No. 289 at 21). The court allowed three narrow categories of requests and part of a seventh, rejected three broader ones and the rest of the seventh, and applied the ordinary Rule 26 analysis rather than any AI-specific rule. It is an interim discovery order entered October 21, 2026 in a live case, not a final or merits ruling, and the record is still developing.

Are my AI chatbot conversations private if I get sued?

There is no categorical answer in this order, and it addresses only civil discovery in one federal case. Under Federal Rule of Civil Procedure 26(b)(1), material that is relevant to a claim or defense and proportional to the needs of the case is generally discoverable unless it is privileged or protected. Whether a particular chat session is protected depends on why it was created, not on the fact that a chatbot was involved.

What is the work-product doctrine, and can it cover a chatbot session?

Rule 26(b)(3)(A) ordinarily shields "documents and tangible things that are prepared in anticipation of litigation or for trial by or for another party or its representative." It is not limited to lawyers, so a party's own materials can qualify, but the test is purpose: the material must have been created principally to assist in anticipated or ongoing litigation. A chat session used to prepare a defense may qualify; one used to draft public posts about the opposing side ordinarily would not.

Does this order force the AI companies to hand over anyone's chats?

No. The order says expressly that "Nothing in this Order compels a nonparty provider to produce information or resolves objections that may be raised by a subpoena recipient." It permits the plaintiff to serve narrowed Rule 45 subpoenas on four unnamed providers. Under Rule 45(d) a recipient may serve a written objection or move to quash or modify a subpoena, including on the ground that it requires disclosure of privileged or protected matter or imposes an undue burden.

Why did the court refuse to make the defendant consent to the subpoenas?

The court denied that oral motion, ECF No. 288, reasoning that Rule 34 requires a responding party to produce responsive nonprivileged materials within its control but "does not expressly require a party to execute an authorization allowing a third party to disclose records directly to an adversary," and that the plaintiff cited no authority supporting the request. Citing Williams v. Carnival Corp., 2020 WL 854809 (S.D. Fla. Feb. 11, 2020), it noted that courts in the circuit generally avoid compelling authorizations where ordinary third-party discovery is available.

What is a privilege log, and why did the order require one?

Rule 26(b)(5)(A) requires a party withholding otherwise discoverable material on privilege or trial-preparation grounds to expressly make the claim and to describe the withheld items in a way that lets the other side assess the claim without revealing the protected content. The court rejected the defendant's blanket work-product objection but allowed document-specific claims, so a log is the mechanism for asserting them.

Why was the plaintiff seeking this material as damages discovery?

Liability had already been resolved: on September 17, 2026 the district court struck the defendant's answer and affirmative defenses and entered default judgment on liability as a Rule 16(f) sanction (ECF No. 261), leaving damages for trial. The plaintiff sought uncapped punitive damages under Fla. Stat. § 768.73(1)(c), which removes the cap where the factfinder finds the defendant had "a specific intent to harm the claimant" and that the conduct did in fact harm the claimant, so evidence of intent and reprehensibility was argued to be relevant.

Is this ruling binding on other courts?

No. It is a discovery order by a United States magistrate judge in the Southern District of Florida, not an appellate decision. The order itself notes that the two 2026 district-court decisions on AI and work product that the parties cited, United States v. Heppner and Warner v. Gilbarco, Inc., establish no categorical rule, and other courts remain free to reach their own conclusions.