Home › Claude Security & Privacy Center for Business Owners › Security questions › After U.S. v. Heppner, how does a firm show its AI use is counsel-directed?

Law firms

After U.S. v. Heppner, how does a firm show its Claude AI use is counsel-directed?

Short answer

A firm shows counsel-directed AI use by making the record show three things Heppner lacked: the work happened in a firm account under commercial terms, a lawyer directed it for a defined matter purpose, and a lawyer reviewed what came out. Heppner was a client using a consumer Claude account on his own, and the court said his outputs were neither privileged nor work product. It did not rule on lawyers using business tools at counsel’s direction. A short checklist and a firm setup make that difference visible. Here is how we set Claude up for law firms.

After U.S. v. Heppner, how does a firm show its AI use is counsel-directed?

What the rules and Anthropic actually publish

United States v. Heppner (S.D.N.Y., Judge Rakoff; oral ruling February 10, 2026, written opinion February 17, 2026): a criminal defendant used a consumer version of Claude, on his own and after retaining counsel, to research the investigation, then sent the outputs to his lawyers. The court held the documents were not protected by attorney-client privilege or work product (Proskauer; Crowell tracker).
The court’s reasons, as summarized by Proskauer and Perkins Coie: Claude is not a lawyer; the consumer privacy terms allowed training and disclosure to third parties, undercutting any expectation of confidentiality; he acted on his own initiative, not at counsel’s direction; and sending the outputs to counsel later did not make them privileged. The court suggested the analysis might differ had counsel directed the use.
Warner v. Gilbarco, Inc. (E.D. Mich., February 10, 2026): a pro se plaintiff’s ChatGPT use did not waive work-product protection, because AI programs are “tools, not persons.” Morgan v. V2X, Inc. (D. Colo., March 30, 2026) also held AI-assisted litigation work can be work product (Crowell tracker). None of these are appellate decisions.
Anthropic Commercial Terms: Anthropic may not train models on Customer Content, and Customer Content is the customer’s confidential information. Anthropic’s “Using Claude for legal work” article notes Heppner turned on consumer terms and recommends commercial terms with a DPA, a lawyer in the loop, verification and documentation.
ABA Formal Opinion 512 (July 29, 2024): competence, confidentiality, communication and supervision duties under Rules 1.1, 1.4, 1.6, 5.1 and 5.3 apply to generative AI use.

The follow-up questions owners ask

Does Heppner mean our lawyers’ Claude work is not privileged?

No. Heppner involved a client, a consumer account and no lawyer involvement. Your lawyers working in a firm account on a matter is a different set of facts, and no court in these decisions ruled against that. Still, these are trial-court rulings, so build your record as if someone will test it.

Under Heppner, what about AI privilege when clients use Claude or other AI on their own case?

That is the Heppner fact pattern, and it is the bigger risk for most firms. Add a line to your engagement letter or intake materials telling clients not to put matter information into AI tools on their own, and to ask you first. If a client should use AI for a task, direct it in writing and say what to do with the output.

What documentation shows counsel-directed Claude use after Heppner without a paperwork burden?

A matter-level note, not a log of every prompt. One line in the file saying the lawyer used the firm’s Claude account for a stated purpose, plus the lawyer’s reviewed work product, is usually enough to show direction and review. Keep the Claude chats inside the matter project rather than in someone’s personal history.

Does Claude Team keep enough of a record to show counsel direction after Heppner?

Team keeps chats in the firm’s account until deleted and the firm can export its data, but it has no audit logs. Audit logs and the Compliance API are Enterprise features. For most small firms, the matter note plus the project history is the record; firms with litigation-heavy practices may want Enterprise for the logs.

What this means for your rollout

Firm account: one firm-owned Claude Team or Enterprise organization under the commercial terms; no personal Free, Pro or Max accounts for matter work.
Matter projects: one Claude project per matter, named to match your matter number, with access limited to the team on the matter.
Written direction: project instructions that state the matter purpose and that the work is prepared at the direction of the responsible attorney.
Documentation: a one-line file note when AI is used on a deliverable, and a client instruction on their own AI use.
Review: a lawyer verifies every citation and conclusion before anything leaves the firm, and the firm’s AI policy says so.
Michael’s two cents

Heppner is a client-behavior case more than a law-firm case, and that is where I see most small firms exposed. I set up the firm account, the matter projects and the project instructions so direction and review are built in, and I help you add the client-facing line. It takes an afternoon, and it means your record tells the right story if anyone asks.

Related services and tools

Claude security for law firmsfirm setup, matter projects and supervision.
Secure Claude implementationconfiguration, policy and training as one project.
Claude usage policythe one-page rule, including the review requirement.
Claude Team vs Enterprisewhen audit logs justify the upgrade.

Get the plan before you buy the seats

The plan choice is the easy part. Configuration, governance and training are where firms get it wrong. The Claude Build Plan is two working sessions with Michael Robichaud, then a written plan: architecture, plan choice with the reasons, your firm’s custom Claude plugin scoped, rollout timeline and cost. A fixed , and the plan is yours to keep.

Get the Claude Build Plan or book a 20-minute intro call.

Related questions

Sources

Michael Robichaud

About the author

Michael Robichaud, Founder of Strategized.ai, sets up Claude for small firms that handle client data: CPA and tax firms, law firms, RIAs, insurance agencies and medical practices. Nearly 30 years in enterprise software sales (RSA Security), digital marketing and technical SEO, and two companies built and sold. Every source on this page was checked against Anthropic’s published documentation and the regulator’s own text.

Book a short call about your firm How a secure setup works

Not legal advice. Confirm regulatory items with your counsel. Anthropic pages change; every source above shows the date we checked it.
Strategized.ai is independent and not affiliated with or endorsed by Anthropic. Claude is a trademark of Anthropic, PBC.