Something cracked open in early 2026. Within the span of weeks, two federal courts examined nearly the same question — does privilege protect generative AI interactions from discovery? — and reached opposite conclusions. That divergence is not a temporary anomaly waiting for appellate resolution. It is, as of now, the operative reality for every litigator and in-house counsel who has fed a confidential document into an AI tool and assumed the output was safe from opposing counsel's reach.
The Rulings That Split the Landscape
The fault line became visible with two decisions issued in February 2026. In United States v. Heppner (S.D.N.Y., Feb. 17, 2026), Judge Jed Rakoff held that documents a criminal defendant generated using a publicly available consumer AI platform were neither attorney-client privileged nor protected as work product. The reasoning was direct: the platform's terms of service explicitly permitted user data to be used for model training, which Judge Rakoff concluded destroyed any reasonable expectation of confidentiality. Without that expectation, the foundational element of privilege — the maintenance of secrecy — simply did not exist.
Days earlier, a court in the Eastern District of Michigan reached the opposite conclusion in Warner v. Gilbarco, Inc. That court reasoned that AI platforms function as tools, not persons, and that routing information through a software tool is categorically different from disclosing it to an adversary. On that logic, using an AI platform to develop legal strategy preserves work-product protection rather than waiving it.
A May 2026 analysis published by the National Law Review confirmed that at least four federal decisions in the first half of 2026 are now collectively sketching what can only be described as an emerging — but deeply unsettled — judicial framework. The train tracks, as that analysis put it, are colliding.
Why the Platform Choice Is Now a Privilege Decision
The Heppner holding introduces a variable that most legal teams have not yet accounted for in their privilege analysis: the terms of service governing the AI platform in use. Consumer-facing AI products — tools deployed through personal accounts or general-purpose subscriptions — frequently reserve broad rights to process and retain user inputs. Under Judge Rakoff's reasoning, those terms effectively function as a pre-litigation disclosure, defeating confidentiality before any opposing party ever files a discovery request.
Enterprise AI deployments, by contrast, typically operate under negotiated data processing agreements that prohibit training on customer data and impose confidentiality obligations. The Warner court's logic maps more comfortably onto that architecture: a tool operating within a sealed contractual environment, with no third-party data rights, looks far less like a disclosure and far more like an internal work product workflow.
The implication is significant. The privilege analysis now begins not with what was communicated, but with where it was communicated. Counsel must treat the selection of an AI platform as a threshold privilege decision, not an IT procurement matter.
The ESI Protocol Gap That Amplifies the Risk
Most Rule 26(f) conferences in active litigation do not yet address AI-generated ESI with any precision. That gap is becoming dangerous. Prompts submitted to a generative AI tool, the outputs returned, and the full conversation logs preserved by the platform may all constitute ESI subject to a discovery obligation — or they may be shielded as work product — depending on the jurisdiction, the platform, and how clearly the parties have addressed the issue in their ESI protocol.
Greenberg Traurig's eDiscovery Watch has highlighted that the absence of explicit AI ESI provisions in case management orders leaves courts and parties without a shared framework for resolving these disputes, creating fertile ground for satellite litigation over discoverability and privilege. Smarsh's analysis of court warnings in this area similarly underscores that judicial patience for ad hoc, mid-case AI governance is limited. Courts are beginning to expect that sophisticated parties will have addressed these questions before they arise.
What In-House Counsel Must Do Now
The divergence between Heppner and Warner does not resolve itself by waiting for circuit-level clarity. It demands immediate governance action across three dimensions.
First, organizations must differentiate between consumer and enterprise AI tools in written retention and privilege policies. The platform determines the privilege exposure, and that determination must be made in advance, not in response to a discovery dispute.
Second, AI usage guidance must be issued to all attorneys and staff involved in matter work — specifying which platforms are approved for use in connection with client matters, under what conditions, and with what documentation requirements for work-product purposes.
Third, every Rule 26(f) conference in active litigation should now include an explicit discussion of AI-generated ESI: what exists, how it was generated, what platform was used, and what privilege claims attach. Opposing counsel and courts are increasingly alert to this issue, and a failure to address it proactively invites exactly the kind of mid-litigation crisis that privilege doctrine is designed to prevent.
How Law & Forensics Helps
Law & Forensics works with litigation teams and in-house counsel to build AI ESI governance frameworks that hold up under judicial scrutiny — from platform assessment and privilege policy design to Rule 26(f) conference preparation and forensic analysis of AI-generated document sets. As courts continue to issue conflicting guidance, having a defensible, documented AI governance structure is no longer optional. Contact Law & Forensics to assess your current exposure and build the protocols your matters now require.