As artificial intelligence becomes increasingly embedded in legal workflows, managing partners and in-house legal leaders must confront a critical threshold issue: whether legal advice created with the assistance of AI is protected by the attorney-client privilege. Under current federal and Florida law, there is no independent or freestanding concept of “AI privilege.” Privilege analysis remains grounded in long-established doctrines that focus on the role of the lawyer, the purpose of the communication, and the preservation of confidentiality—not on the novelty of the technology used. Nothing in this analysis should be construed as asserting the existence of any independent or freestanding “AI privilege.” Any claim of protection arises solely, if at all, under traditional attorney-client privilege and work-product doctrines as recognized under federal and Florida law.
Privilege Analysis Begins With the Human Lawyer
Under federal common law and Florida law, the attorney-client privilege protects confidential communications between a client and a licensed attorney made for the purpose of seeking or providing legal advice. See Upjohn Co. v. United States, 449 U.S. 383 (1981); Fla. Stat. § 90.502. The purpose of protecting the attorney-client communications is to allow clients to speak freely with their counsel and to allow attorneys to render advice, guidance and commentary, all without the fear that their words may be used in court against the client. These frameworks do not extend privilege to communications generated by non-lawyers acting independently of an attorney’s professional judgment. Accordingly, advice produced directly by a generative AI system and delivered to a business user or client—without substantive attorney involvement—cannot be privileged. An AI system is not admitted to practice law, is not subject to ethical rules, and cannot form an attorney-client relationship. Simply, attorney-client privilege does not attach to communications generated by an artificial intelligence system acting independently of a licensed attorney, nor to raw AI outputs that have not been reviewed, modified, and adopted through the exercise of professional legal judgment by counsel.
The analysis shifts, however, when AI is used by a licensed attorney as an internal support tool. Federal and Florida courts have long recognized that the involvement of non-lawyers does not destroy privilege where those individuals or resources are acting at the direction of counsel, in furtherance of the provision of legal advice, and within a confidential relationship. See, e.g., United States v. Kovel, 296 F.2d 918 (2d Cir. 1961); Fla. Stat. § 90.502(1)(c). In this context, generative AI may be analogized to other subordinate legal resources—such as paralegals, junior attorneys, or outside service providers—whose work is filtered through and ultimately owned by the supervising lawyer. The AI platform used, however, may be scrutinized to determine if this analogy should be applied; if sensitive information is input into AI platforms that provide no guaranty of privacy, data control, or limited retention—like a supervising lawyer discussing legal matters with their waiter at lunch—the information may still be discoverable and admissible in court.
The party asserting attorney-client privilege bears the burden of establishing that any AI-assisted communication was made for the purpose of seeking or rendering legal advice, under the direction and supervision of counsel, and within a preserved confidential relationship. Contemporaneous documentation reflecting attorney review and approval materially supports that showing.
Where a qualified attorney uses AI to assist with research, drafting, or issue spotting, then independently evaluates the output, applies professional judgment, and adopts or revises the analysis before communicating advice to the client, the resulting communication may qualify for attorney-client privilege. The critical point is that the legal advice must remain the attorney’s own. Privilege, if it applies, attaches to the lawyer’s communication—not to the AI system or its raw output. Separately and independently, materials prepared by or at the direction of counsel with the assistance of AI in anticipation of litigation may qualify as protected attorney work product, including opinion work product reflecting legal theories, mental impressions, or litigation strategy, even where attorney-client privilege is disputed.
Notwithstanding, confidentiality remains the most significant risk factor in this analysis. From both a federal and Florida perspective, the use of public or consumer-grade AI platforms raises substantial privilege concerns. Many such systems reserve rights to store, review, or reuse user inputs, which can defeat any reasonable expectation of confidentiality and support an argument that privilege has been waived. Law firms should therefore treat unrestricted AI platforms as inappropriate for handling client facts, legal strategy, or draft advice.
Prudent firms and legal departments should implement disciplined governance around AI usage. This includes selecting enterprise or closed-system AI tools with robust contractual safeguards, requiring documented attorney supervision of all AI-assisted legal work, and training lawyers and staff on privilege risks and ethical boundaries. In sensitive matters, maintaining records that demonstrate attorney review and approval may further support a claim that the communication reflects protected legal advice rather than unvetted machine output.
Conclusion
Ultimately, artificial intelligence can be integrated into legal practice without forfeiting attorney-client privilege, but only if firms remain anchored to the foundational principles of federal and Florida law (and any other applicable law). Privilege flows from the lawyer’s role, judgment, and duty of confidentiality—not from the technology used. Without careful oversight and clear internal controls, AI adoption may create avoidable exposure to privilege challenges and waiver arguments that law firms cannot afford to overlook.
