Attorney-Client Privilege and the Work Product Doctrine in the Age of Legal AI
How attorney-client privilege and the work product doctrine applies to enterprise legal AI, what the first wave of 2026 court opinions tells us, and how Harvey is built to support the confidentiality discipline that privilege requires.
Written by John LaBarre, Chief Legal Officer, Harvey
Download White PaperSummary
As AI becomes part of everyday legal work, lawyers face a familiar question in a new context: how do we use new technology while protecting confidential communications and litigation materials? The first US court decisions addressing AI and these protections offer early guidance, with close attention to who uses the technology, how it operates, and how information is protected.
Written by Harvey’s Chief Legal Officer, this paper examines how attorney-client privilege and the work product doctrine apply to enterprise legal AI. It explores the emerging case law, how Harvey’s design supports confidentiality, and the governance questions that arise as AI moves from analysis to action.
Key Takeaways
- Nearly every prior technology shift has begged the question of whether privilege has been preserved by asking one question: was the client’s confidence exposed to an outsider in a way inconsistent with the protection? Large language models pose that question again.
- The first 2026 court opinions distinguish consumer AI tools from enterprise legal platforms. Two courts adopted the “tool” framing and held work product protection intact; the one court that denied protection did so on facts bearing no resemblance to a lawyer using enterprise legal AI.
- The recurring analytical questions are: who is using the AI, what kind of platform is involved, and whether AI functions as a tool or a third party. Outcomes turn on reasonable steps to preserve confidentiality.
- Harvey is built for lawyers, not general consumers. Beyond enterprise agreements with stringent contractual confidentiality protections, we offer authenticated workspaces, no training on customer data, no retention beyond ephemeral processing, no human review, and regional data processing in customer-selected regions.
- Privilege is a discipline practiced by lawyers and clients in context, not a guarantee any vendor can provide in the abstract. But well-designed legal technology (like Harvey) can make that discipline easier to maintain.
- As AI enters the agentic era, governance must exist at the point of action: controls over which actions the system may take autonomously, which require human review, and which systems the tool can reach.