Before You Prompt: AI, Privilege, and Confidentiality for Florida Lawyers

Privilege and confidentiality questions are fact-specific. This post is informational, not legal advice. A version of this article first appeared as a member article for the Business Law Section of The Florida Bar; this version adds the Massachusetts Shealy decision.

Your client sends over a messy set of deal documents or litigation materials and you do what increasingly feels normal: you paste them into a generative AI tool and ask it to find the problems. You just saved an hour. Depending on the tool and its terms, you may also have handed a third party your client’s confidential information, and, if litigation follows, given your adversary an argument that your analysis is fair game in discovery. For Florida business lawyers, that tension is no longer hypothetical.

Florida is one of the few states with guidance directly on point. Florida Bar Ethics Opinion 24-1 (Jan. 19, 2024) tells lawyers they may use generative AI, and Florida has since amended its competence and verification rules to match. The question is no longer whether Florida lawyers will use these tools. It is whether they can use them without forfeiting privilege or breaching confidentiality.

What the courts are starting to say

No Florida appellate court has yet ruled on whether AI-assisted work keeps its privilege or work-product protection, so the early roadmap comes from out-of-state decisions issued almost entirely in 2026.

The cautionary case is United States v. Heppner, 820 F. Supp. 3d 292 (S.D.N.Y. 2026). A criminal defendant used a consumer AI tool on his own, without his lawyers directing the work, to draft defense analysis he later shared with counsel. The court held the materials were neither privileged nor work product. The consumer tool’s terms let the provider keep and train on what he typed and share it with outside parties, so he had no reasonable expectation of confidentiality, and feeding information to the tool was treated like sharing it with any other third party.

The decisions in civil cases are more protective, but fact specific. In Warner v. Gilbarco, Inc., 820 F. Supp. 3d 629 (E.D. Mich. 2026), Tate Group Automotive, LLC v. Legacy Automotive Capital, LLC, No. 25-BC11B-0020 (Tex. Bus. Ct. 11th Div. June 3, 2026), and Assini v. Hayward, 2026 NY Slip Op 26086 (Sup. Ct. Nassau Cty. June 4, 2026), the courts held that a party’s AI-assisted litigation work can be protected work product, and that using a public AI tool does not, by itself, waive that protection. Their reasoning is that work product is waived by disclosure to an adversary or by conduct likely to put the material in an adversary’s hands, and typing into an AI tool is not automatically that. Assini adds an important procedural twist: the subpoena was directed to OpenAI itself, and the court still quashed it after finding Morgan persuasive. Tate is worth a Florida practitioner’s attention because it involved a represented party, a non-lawyer business principal, and ordinary commercial litigation, the setting most business lawyers work in.

The newest decision cuts the other way, and it lands closest to home for lawyers with represented clients. In Shealy v. Seaside Investments, LLC, No. 2684CV00799-BLS2 (Mass. Super. Ct. June 16, 2026), a business dispute over the sale of a Boston gas station, the plaintiff received a default letter and forwarded it, along with the deal documents, to his romantic partner. At his request she uploaded them to ChatGPT, generated a response letter and other documents, and emailed the output back to him. His lawyers were never involved. The court compelled production of all of it, queries and output alike. The partner was not a “representative” under the work product rule (the rule’s list runs to a party’s attorney, consultant, surety, indemnitor, insurer, or agent), and the court declined to stretch the doctrine to cover a friend or partner with no substantive connection to the litigation. It distinguished Warner and Morgan as cases about pro se litigants who were both party and advocate, and followed Heppner: AI materials generated without counsel’s direction do not reflect counsel’s strategy. The court also held that ChatGPT’s output is not opinion work product, because “AI is a tool not a person.”

Two themes run through these cases. Privilege is fragile, because disclosure to a third party can waive it unless the disclosure fits within an agency, service-provider, common-interest, or similar protected framework, and an AI vendor may be treated as a third party. Work product is sturdier, but courts are policing the inputs. In Morgan v. V2X, Inc., No. 1:25-cv-01991, 2026 WL 864223 (D. Colo. Mar. 30, 2026), the court required the litigant to identify the AI tool used with confidential information and barred confidential material from consumer platforms that lacked specific contractual safeguards. Shealy adds a third theme: courts are policing who directs the prompting. For a represented party, protection may turn on whether the AI work was done at counsel’s direction. Self-help AI analysis by the client, or by someone the client recruits, sits outside the shield.

What Florida’s rules require

Opinion 24-1 conditions AI use on duties Florida lawyers already owe, and Florida grounds those duties in its own law. Although the opinion is advisory, the underlying duties are not. The evidentiary lawyer-client privilege under Section 90.502, Florida Statutes, is narrower than Rule 4-1.6, Rules Regulating The Florida Bar, which reaches all information relating to the representation. The practical consequence is that a disclosure to an AI tool can survive a privilege challenge and still be an ethics violation. Opinion 24-1 tells lawyers to research a tool’s policies on data retention, data sharing, and self-learning before use, and to obtain the client’s informed consent before putting confidential information into a third-party tool, consistent with Florida’s earlier guidance on vetting cloud and outside technology providers. See Fla. Ethics Op. 12-3 (Jan. 25, 2013).

Competence is the next duty. In 2024 the Florida Supreme Court amended the comment to Rule 4-1.1 to require lawyers to understand the benefits and risks of relevant technology, including generative AI. In re Amends. to Rules Regulating the Fla. Bar, Ch. 4, 393 So. 3d 137, 139 (Fla. 2024). Florida had already made technology competence concrete with a three-hour technology CLE requirement under Rule 6-10.3. A lawyer who cannot say what a tool does with an input is not yet competent to put a client’s confidences into it. Supervision is the related duty: Opinion 24-1 treats a generative AI tool much like a nonlawyer assistant under Rule 4-5.3, so the lawyer stays responsible for the result.

Florida now backs the verification duty with a filing rule. Effective June 15, 2026, amended Rule 2.515(d)(2) requires whoever signs a court filing to represent that the legal authorities cited exist and are accurately cited, and it authorizes sanctions, after notice and an opportunity to be heard, for filings that fall short. In re Amendments to Fla. R. Gen. Prac. & Jud. Admin. 2.515, No. SC2026-0673 (Fla. May 28, 2026). A fabricated citation from an AI tool can now be sanctionable in every Florida court.

Practical steps

A few measures address most of the risk:

Classify the task before you prompt. Generic drafting or brainstorming may be fine for a public tool, but client facts, deal strategy, board advice, diligence, and litigation analysis call for a controlled environment.
Use business or enterprise-tier tools whose contracts bar training on your data and give you deletion rights. Keep client work out of consumer chatbots.
Get the client’s informed consent before any tool will touch confidential information and keep privileged communications and produced discovery out of public tools entirely.
Tell clients, at the start of a dispute, not to run the dispute documents through AI on their own and not to hand them to a spouse, partner, or friend to do it. After Shealy, that do-it-yourself analysis may be produced to the other side.
Document that AI use was at counsel’s direction in anticipation of litigation and keep a record of the tool used and what was entered. Some courts are now ordering that disclosure.
Verify every output before it is filed or sent. A confident answer from a tool is still your work to check, and in Florida a bad citation is sanctionable.

Where this is heading

Florida courts will reach the privilege questions before long, and the out-of-state rulings suggest they will protect AI-assisted work a lawyer directs and controls while declining to protect careless consumer use. Shealy sharpens the line: the same doctrine that shields counsel-directed AI work leaves a client’s freelance prompting exposed. Opinion 24-1 and Florida’s amended rules already point the same way. The tool can save the hour; whether it costs the privilege depends on what the lawyer does before pressing enter.