Client Alert – AI Tools and Trademark Practice: What You Don’t Know Can Hurt You

In light of United States v. Heppner, 2026 U.S. Dist. LEXIS 32697 (S.D.N.Y. Feb. 17, 2026)

 

A February 2026 federal court ruling answers a question that trademark owners and their advisors have been asking since AI tools became ubiquitous: are conversations with a generative AI platform protected by attorney-client privilege or the work product doctrine? The answer is no — and the reasoning applies directly to three tasks that trademark owners increasingly hand off to AI.

 

The Heppner Ruling

In United States v. Heppner, the judge ruled that documents a defendant prepared using Anthropic’s Claude — outlining his legal defense strategy in anticipation of a federal criminal prosecution — were not protected by attorney-client privilege or the work product doctrine. The court ordered them produced to the government in full.

The court relied on three grounds. First, Claude is not an attorney, and no attorney-client relationship can exist with an AI platform. Second, Anthropic’s privacy policy permits voluntary disclosure of user inputs to third parties, eliminating any reasonable expectation of confidentiality. Third, documents prepared without attorney direction — even in genuine anticipation of litigation — are not protected work product when voluntarily transmitted to a third-party platform that may share them. This was a question of first impression, and the ruling is not nationally binding. But the logic is sound and its application to trademark practice is direct.

 

Three High-Risk Uses of AI in Trademark Matters 

1. Trademark Clearance Searches

Inputting a proposed mark, product description, or brand strategy into a public AI platform discloses that information to a third party under terms that permit further disclosure. There is no privilege or work-product protection. Your analysis of potential conflicts — which marks you identified as risky, which you dismissed, and how you reasoned through the decision — will likely be discoverable in any subsequent infringement dispute. Prompts that acknowledge a competing mark while asking for reasons to proceed anyway can establish bad faith or willful infringement.

 

2. Evaluating Infringement Claims and Defenses

Asking AI to assess a cease-and-desist letter or an infringement claim creates a written record of your legal theories, known weaknesses, and vulnerability assessment. A client’s own litigation-oriented analysis may, in the right circumstances, qualify as ordinary work product — but that protection is forfeited when you transmit the analysis to a public AI platform. No amount of attorney oversight cures that forfeiture: the problem is the platform’s third-party disclosure terms, not the absence of an attorney. If you also incorporate your attorney’s advice into the prompt, that information faces the same exposure and potentially risks waiving the attorney-client privilege.

 

3. Reviewing Settlement Offers and License Drafts

Submitting a settlement agreement or license draft to a public AI discloses your negotiating position, your walk-away terms, and your valuation of the dispute — all of it potentially discoverable if litigation follows. Many trademark settlement agreements also contain confidentiality clauses that prohibit third-party disclosure; submitting those documents to a public AI platform may constitute an independent breach of the agreement, exposing you to a separate damages claim. Federal Rule of Evidence 408, which limits use of settlement communications at trial, does not shield them from discovery.

 

Does a Private AI System Fix the Problem?

Partially — and meaningfully so for some of these risks. Enterprise AI systems such as Microsoft Azure OpenAI or Claude for Enterprise can be configured by contract so that the vendor does not retain, train on, or disclose your inputs. That eliminates the voluntary third-party disclosure problem. A litigation-oriented analysis prepared on a properly configured closed system, in genuine anticipation of litigation, may retain ordinary work product protection in many jurisdictions — though this is fact-specific and should not be assumed without legal guidance.

What a closed system cannot do is create attorney-client privilege. That still requires an attorney’s direction and involvement. And the absolute prohibition remains: no public AI platform, under any circumstances, for any of the tasks described above. Attorney oversight does not cure a public platform’s disclosure terms.

 

A Practical Framework

****Never use a public AI platform for trademark clearance, infringement analysis, or settlement review — under any circumstances, with or without attorney involvement. For closed enterprise systems, counsel should direct the analysis, and output should be integrated into privileged work product rather than treated as a standalone client document. Before submitting any existing agreement to an AI platform, check it for confidentiality provisions.****

 

This alert reflects the law as of March 2026. The Heppner decision is persuasive authority from the Southern District of New York — not binding nationally — and this area will continue to develop. Please reach out with questions about your specific practices or jurisdiction.

 

Contact – We regularly assist corporate legal departments in developing AI governance frameworks, reviewing enterprise AI vendor agreements, and auditing existing contracts for AI-related confidentiality exposure.

Please contact: Lawrence R. Robins  |  Partner, Chair — Brand Management Practice Group [email protected].

If you are interested in a more in-depth analysis of this decision please reach out.