Bring counsel in before the prompt becomes a problem.
AI can help your team work through sensitive questions. Whether that work is protected depends on the circumstances. Heppner shows why legal purpose, counsel’s involvement, and confidentiality deserve attention before the conversation begins.
Privilege AI brings risk detection, counsel-directed work, and matter-level records into one legal risk governance platform.
What happened in Heppner?
Bradley Heppner used publicly available Claude to explore potential defenses while under investigation. His attorneys had not directed that work. After records of the exchanges were seized, he asserted attorney-client privilege and work-product protection, including on the basis that he had shared the materials with counsel.
The court rejected those claims. Its reasoning focused on the absence of an attorney-client communication, the confidentiality implications of the provider policy before it, and the lack of counsel’s direction or strategy in the creation of the materials. Sending the documents to counsel afterward did not make the original exchanges protected.
For a business, this is a concrete reason to examine the process around AI: who is directing the work, what purpose it serves, what information is being disclosed, and under which terms.
Read the facts and reasoning in the court memorandum, pp. 3–12What the ruling does — and does not — tell us
An AI answer is not an attorney-client relationship.
A sensitive subject or a legal-sounding response does not, by itself, establish privilege. Counsel’s role and the purpose of the communication matter.
Confidentiality needs a real foundation.
Provider terms, access, disclosure rights, data use, and the way a tool is configured all deserve review. The court examined the policy applicable to the use before it; this page does not treat every AI product or contract as identical.
Legal review afterward may come too late for the earlier exchange.
Moving work into a legal matter can improve the process going forward. It does not automatically undo an earlier disclosure or change the status of pre-existing material.
Counsel-directed use remains a fact-specific question.
The court discussed the possibility of an agency argument where counsel directs AI use. It did not create an automatic safe harbor. Other courts have reached different results in different work-product settings.
Courts are still defining how privilege and work-product protections apply to AI. The result depends on the facts, the purpose of the work, the people involved, the platform’s terms, and the applicable law.
The business problem starts before a lawsuit
An executive asks how to respond to a regulator. HR drafts a disciplinary letter. A sales leader analyzes a disputed contract. A finance team summarizes a restricted data room. Each task can mix ordinary business work with legal advice, confidential information, or preparation for a dispute.
The important moment is often the first prompt. Your team needs a practical way to recognize a legal question and bring it to counsel, with clear rules for the information it can use.
One platform, three connected roles
Surface the questions that need legal review.
Watch brings supported company-account AI alerts and Agent detections into a review queue for legal. Teams can review context, identify questions requiring counsel, and route work into a matter. A detection is a suggestion for review, not a finding of misconduct.
Explore Privilege WatchWork the question with counsel.
Organize AI-assisted legal work around a matter, with attorney roles, a reviewed basis for the session, and records of the model, citations, and participants. Give counsel a structured place to direct the work and assess the claim of protection.
Explore the counsel-directed workspaceAdd visibility where work happens.
On enrolled, supported workstations, the Agent surfaces potential policy concerns and sends evidence to Privilege Watch for review. Device coverage and detection context help legal and security teams understand where follow-up is needed.
Explore endpoint visibilityDetection, legal review, and a documented matter work together. None of them retroactively makes an earlier AI disclosure privileged.
Build a process counsel can explain
Recognize the legal question.
Identify work involving disputes, investigations, employee issues, transactions, or other sensitive legal matters. Escalate it under your organization’s policy.
Establish counsel’s role and the purpose.
Have counsel determine whether legal advice or qualifying litigation preparation is involved, who should participate, and how AI may assist.
Review the information and the environment.
Confirm permitted inputs, access, provider terms, retention, and any contractual or court-ordered restrictions before information is uploaded.
Work within the matter.
Use the approved workflow to collaborate, review sources, and document the model and participants. Counsel remains responsible for legal judgments and review of AI output.
Preserve and assess the record.
Follow applicable retention and legal-hold requirements. Give counsel the information needed to assess confidentiality, protection, and any response to a request for production.
Related governance concerns: the same operating process can help teams review framework compliance and respond to insider threats. Those concerns have their own requirements; Heppner addresses privilege and work product in the circumstances of that case.
Relevant wherever sensitive decisions meet AI
More industry applicationsProfessional services · Manufacturing · Retail and e-commerce · Insurance · Construction and real estate
Questions legal teams are asking
Does Heppner mean every AI conversation is discoverable?
No. The decision addressed particular exchanges in a criminal case. Whether material must be produced depends on applicable procedure, the request or legal process, relevance and other limits, and any valid protection. Heppner is a reason to review your workflows, not a universal rule about every AI chat.
Is attorney-client privilege the same as work product?
No. Attorney-client privilege generally protects qualifying confidential communications for legal advice. Work-product protection applies to qualifying litigation preparation and has different requirements. The Heppner court rejected both on the facts before it.
Does having a lawyer in an AI session guarantee protection?
No. Counsel’s involvement matters, but so do purpose, confidentiality, participants, disclosure, and applicable law. A session label or attorney invitation cannot establish all the necessary facts by itself.
Can sending an existing chat to a lawyer make it privileged?
It does not automatically protect the original chat. In Heppner, later sharing did not retroactively protect the AI materials. A separate confidential communication with counsel requires its own analysis. Ask counsel how to handle existing records and preserve them as required.
Do an enterprise account or no-training terms settle the question?
No. They may be relevant to confidentiality, but counsel must examine the actual service terms and configuration, including access, retention, disclosure, and permitted uses. Information security and legal privilege answer different questions.
What does the audit record contribute?
It can help counsel establish what happened: the matter, participants, model, sources, and recorded workflow. A tamper-evident record supports review; it does not prove that every input was accurate or that a court will uphold a claim of protection.
What happens when Watch flags a conversation?
Legal reviews the context and decides the next step. A flagged question can be routed into a counsel-directed matter. The alert does not determine whether a violation occurred or whether earlier material is privileged.
Is this page legal advice?
This is general information about a court decision and the workflows Privilege AI is designed to support. Privilege AI is software, not a substitute for your lawyer. Your counsel should assess the facts, jurisdiction, applicable rules, and approved tools for your matter.
Make the next sensitive AI question a conversation with counsel.
Walk through how your organization can identify risky AI use, bring counsel into the work, and maintain the matter record. Start with one real workflow — an employment issue, contract dispute, investigation, or transaction.
No software guarantees attorney-client privilege or work-product protection. Protection depends on the facts and applicable law. This page discusses the February 17, 2026 Heppner memorandum; courts continue to address AI use in different settings.
- United States v. Heppner, No. 25-cr-503 (JSR), ECF 27 (S.D.N.Y. Feb. 17, 2026). Read the February 17, 2026 memorandum, especially pp. 3–12 (searchable mirror).
- Other AI work-product decisions: Warner v. Gilbarco, February 10, 2026, pp. 9–12; and Morgan v. V2X, March 30, 2026, pp. 7–11 and 14–16. These cases involve different facts and illustrate the developing treatment of AI-assisted litigation preparation.
Research reviewed · October 31, 2026