Can In-House Counsel Spell “Privilege” Using AI?

Key Takeaways: AI is not a lawyer—or a vault. Sharing privileged information with many artificial intelligence (AI) platforms may constitute disclosure to a third party, triggering waiver of the attorney-client privilege, because those tools can store prompts, use them for training, or share data externally. The email era offers a warning. Just as rules for email confidentiality began to develop in the 1990s—only to evolve over two decades—new case law around AI is just beginning to emerge. Initial signals suggest courts are willing to compel disclosure of AI-generated exchanges, making proactive caution essential. Legal oversight is the answer. In-house counsel should lead the development of AI policies, ensure that legal questions go to lawyers rather than chatbots, and document any AI use within privileged communications.

Artificial intelligence (AI) is everywhere.

Lawyers now use it with increasing regularity, and so do staff, colleagues, and clients. We are harnessing generative AI for simple tasks, like transcribing meetings, and for more complex work, including research, drafting, and preparing presentations and articles, like this one.[1]

For in-house counsel, this reality adds a fresh complication to attorney-client privilege. AI stress-tests every condition required to preserve privilege, so corporate counsel must weigh its use against the risk of waiver.

A Quick Refresher on Attorney-Client Privilege

A mechanism for shielding information might seem inapposite to a justice system that relies on evidence. Yet the attorney-client privilege is essential because it promotes transparency through full, frank disclosure between clients and lawyers. This is especially important for in-house counsel needing to provide legal advice within a business setting.

In-house lawyers, however, often provide both legal and business advice to their sole client. Corporate counsel’s business advice is not privileged, even when provided confidentially. Protection applies only to communications that are primarily legal in nature, so not every discussion that includes a lawyer is privileged.

When an entity uses legal advice to make business or employment decisions, attorney-client privilege may still protect underlying communications between counsel and the organization, including employees, as long as they remain confidential and have a legal purpose. But courts construe privilege narrowly, and waiver, sometimes inadvertent, can occur easily, including through disclosure to a third party.

Waiver and the Specter of AI

Technology makes work easier, but it can also complicate things in unforeseen ways. Uninformed or insecure use of technology can lead to unintended waiver of the attorney-client privilege. Even though inadvertent waiver can be rectified (sometimes), no one wants to subject themselves or their client to extra discovery motion practice over communications that should have remained protected.

Today, we generally accept email as sufficiently confidential to maintain privilege. But that was not always obvious. After email became standard in the 1990s, legal commentators questioned whether messages sent through providers like America Online (AOL) provided a reasonable expectation of privacy and qualified for the attorney-client privilege.[2] Case law that followed offered cautionary tales—forwarding messages carelessly, copying unnecessary recipients, or dropping counsel from a thread can waive privilege. 

From that interplay of comments and cases and the reality of lawyers and clients communicating through such platforms came today’s working rule: Email can be privileged if the subject is predominantly legal in nature, the parties to the communication are necessary participants within the privileged relationship, and sufficient precautions are taken to protect the privilege.

Now, AI poses the next challenge, presenting significant risks to the attorney-client privilege in the corporate environment, with potentially dire consequences for discoverability in litigation.

Red Light, Green Light, 1, 2, 3 . . .

Whether AI-related communications can be privileged is a fairly new question already surfacing in litigation. The few courts to address it have signaled willingness to compel disclosure of AI-generated exchanges.[3] Thus, while organizations everywhere are greenlighting AI-fueled processes for meeting transcription, spreadsheets, and drafting, in-house counsel should consider providing legal advice about those business decisions.

AI use may be inevitable, but heightened care is crucial to avoid compelled disclosure of legal strategy and other confidential exchanges. Several features of AI use can disrupt attorney-client privilege:

• AI is not a lawyer, so communications with it are not inherently privileged.
• AI is not an employee, so disclosure to it may count as “going outside” of the company.
• AI exchanges are often not confidential. Many platforms store prompts, use them for training others, or share data with third parties.
• AI may be helpful, but not necessary, for effective lawyer-client consultation.

Moreover, the conversational nature of prominent AI tools makes it easy to forget that they often aren’t private spaces. Closed, enterprise systems may mitigate risks after sound technical and legal review.

To avoid inadvertently waiving the attorney-client privilege, consider these cautionary yellow-light points:

• Disclosure to AI is disclosure to a third party unless the system is appropriately secured. Train users to think about confidentiality and technology selection. Information that may be acceptable in a closed enterprise tool may be inappropriate for an open platform.
• WHY < WHAT & WHO: Privilege turns on structure, not intent. The reason for using AI does not control the analysis; communications with AI are judged under the same privilege rules as other disclosures.
• Governance is foundational. Give stakeholders proactive, pragmatic, and up-to-date guidance, policies, and training.
• Unsupervised AI use creates big risks. Decisions about which tools to deploy, for whom, and for what purposes are business choices that should be informed by legal judgment.
• Counsel-directed, secure use may preserve attorney-client privilege. Legal oversight must be clear. Insist that legal questions go to lawyers, not chatbots. Any AI used within lawyer-client communications should be documented as necessary, much like the use of a translator, accountant, or other expert intermediary.[4]

Before sharing information with AI, every stakeholder should ask: Would I be comfortable producing this in litigation? If the answer is NO, that’s a red light—or at least a reason to call a lawyer to ask for directions before you run over the privilege.

ENDNOTES

[1] AI did not, in fact, produce this article. But our colleague Elizabeth A. Ledkovsky, a Staff Attorney at Epstein Becker Green, contributed and—full disclosure—used integrated AI tools to edit the draft.

[2] See, e.g., Ben Delsa, E-Mail and the Attorney-Client Privilege: Simple E-Mail in Confidence, 59 La. L. Rev. (1999). Available at: https://digitalcommons.law.lsu.edu/lalrev/vol59/iss3/10; Janna Fischer, Big Boss is Watching: Circumstances Under Which Employees Waive the Attorney-Client Privilege by Using E-Mail at Work, 12 Colo. Tech. L.J. 365 (2014). Available at: https://scholar.law.colorado.edu/ctlj/vol12/iss2/7

[3] For example, United States v. Heppner, No. 25-cr-00503-JSR (SDNY, Feb. 17, 2026). See also: Elizabeth X. Guo, Blog Essay: United States v. Heppner, Harvard L. Rev. (Mar. 23, 2026). Available at: https://harvardlawreview.org/blog/2026/03/united-states-v-heppner/

[4] See, e.g., United States v. Kovel, 296 F.2d 918 (2d Cir.1961).