Attorney-Client Privilege (AI Context)

Attorney-client privilege is the oldest evidentiary privilege recognized in Anglo-American law, protecting confidential communications between a client and attorney made for the purpose of obtaining or providing legal advice from compelled disclosure in legal proceedings. The privilege belongs to the client, who may assert or waive it; its core rationale is that clients must be able to communicate fully and frankly with their lawyers without fear that those communications will be used against them.

In the AI context, attorney-client privilege intersects with legal technology in several ways. Most immediately: when a lawyer uses an AI tool to process, analyze, or draft materials involving confidential client communications, those communications pass through the AI vendor's infrastructure. The critical legal question is whether that transmission to a third-party AI system destroys or risks the privilege. The prevailing analysis treats AI vendors as functional agents of the attorney—analogous to outside consultants, contract lawyers, or other service providers to whom privileged material is disclosed in the course of representation—provided the vendor relationship is appropriately structured.

The structural requirements for preserving privilege when using AI tools mirror those for other third-party service providers: the vendor must be engaged to assist with legal representation, vendor access to privileged material must be limited to what is necessary for the engagement, the relationship must be governed by confidentiality obligations, and the vendor must not use privileged materials for purposes beyond serving the representation. Vendors that use customer inputs to train their AI models without consent—or that share data across customers—present the most significant privilege risk.

Privilege protection is a foundational requirement in legal practice. Attorneys who use AI tools without understanding how those tools handle client communications may inadvertently create privilege risks that are difficult to remedy after the fact. If an AI vendor's data handling practices are characterized as a disclosure to a third party outside the privilege—rather than as use of an agent—privilege over processed communications could be waived.

The practical risk is heightened in litigation contexts. Opposing counsel aware that an attorney has used AI tools may seek discovery of AI system inputs, outputs, and processing logs on the theory that disclosure to the vendor waived privilege. While courts have generally supported the agent theory for legal AI tools with appropriate contractual protections, the case law is still developing and outcomes are not guaranteed.

Lawyers also face privilege analysis when they advise clients on AI-related matters. Corporate clients deploying AI systems may seek legal advice about those systems; the privilege protecting that advice must be carefully maintained. In-house counsel using AI tools to draft legal memoranda face the same privilege-preservation questions as outside counsel—their communications carry attorney-client privilege when made in a legal capacity, but only if treated accordingly.

Enterprise legal AI vendors have invested significantly in privilege-preserving data architectures. Harvey, Luminance, and CoCounsel operate under enterprise terms that prohibit using customer inputs for model training without consent, maintain strict data isolation between customers, and provide contractual confidentiality commitments that support the agent relationship characterization.

Zero-retention policies—where the AI vendor does not store user inputs or outputs beyond the session—represent the strongest structural protection for privilege. Some enterprise legal AI deployments offer this option, though it often involves tradeoffs with features like conversation history and workflow continuity. Private LLM deployments—where the model runs on infrastructure controlled by the law firm or their cloud environment—address the third-party disclosure concern most directly by keeping privileged material within the firm's own infrastructure.

From a documentation standpoint, attorneys using AI for privileged matters should ensure that vendor contracts include appropriate confidentiality provisions, confirm that the vendor's data handling practices match the contractual commitments, and be prepared to articulate the agent relationship theory if privilege is challenged. The audit logs generated by enterprise AI platforms—which document what data was processed, by whom, and for what matter—can support privilege arguments by demonstrating that AI use was within the scope of a specific legal representation.