Privilege and AI Tools

Short Answer

Privilege depends on a protected relationship, legal-advice purpose and confidentiality. A consumer chatbot is not counsel. In United States v. Heppner, the Southern District of New York rejected privilege and work product for a defendant’s AI research undertaken without his lawyers’ direction and later shared with them. The opinion does not decide every lawyer-directed or enterprise AI workflow. Counsel’s prompts can reflect work product, but reliance on selected prompt-output pairs can create waiver questions. Texas privilege, waiver and work-product rules must be applied to the actual access rights and purpose; a product’s enterprise label does not establish protection.

Which Laws Apply

Texas AI-specific: none.

Generally applicable Texas law: Tex. R. Evid. 503 (lawyer-client privilege) and 511 (waiver); Tex. R. Civ. P. 192.5 (work product); Texas Disciplinary Rule 1.05 (confidentiality); Ethics Opinion 705.

Federal: federal common law of privilege; Fed. R. Civ. P. 26(b)(3); Fed. R. Evid. 502.

Heppner

After receiving a grand jury subpoena and retaining counsel, the defendant used the free consumer version of an AI service to research the government’s investigation, entered information he had learned from his lawyers, generated reports outlining defense strategy and later sent them to counsel. The government moved for a ruling that the materials were not protected. Judge Rakoff ruled from the bench on February 10, 2026 and issued a written opinion on February 17, 2026.

The court reasoned that the AI tool is not an attorney; that the provider’s terms allowed it to use inputs and outputs and disclose them, so the defendant had no reasonable expectation of confidentiality; that sending the outputs to counsel afterward did not make them privileged; and that work product did not apply because the defendant did not prepare the materials by or at the direction of counsel. The court noted that a counsel-directed arrangement might be analyzed differently but did not decide it.

Lawyers’ Prompts as Work Product

The May 23, 2025 order in Concord Music Group v. Anthropic, ECF 377, applied work-product analysis to pre-suit prompts and outputs. It distinguished the materials relied on in the litigation from unrelied-on materials and rejected a categorical waiver of the latter. The order refers to Tremblay’s similar analysis. It is a district-court discovery ruling, not binding Texas precedent. A lawyer should preserve the purpose and direction of the work and identify exactly what the party has put at issue.

Applying This in Texas

Texas privilege law, like federal law, turns on confidentiality and the purpose of the communication. A client who pastes privileged advice into a consumer chatbot risks a waiver argument. A lawyer who uses an enterprise tool under terms that bar training and limit access is in a stronger position, and Ethics Opinion 705 tells Texas lawyers to understand a tool’s confidentiality practices before using it with client information.

Practical Steps

Tell clients, in engagement letters or early advice, not to put privileged communications or case facts into consumer AI tools.

Use tools with written confidentiality, no-training and access commitments for client work.

Document when AI use is directed by counsel for litigation.

Treat prompts relied on in pleadings or expert work as potentially discoverable.

What Is Unsettled

Whether enterprise AI terms preserve confidentiality for privilege purposes; whether a Kovel-type arrangement can extend privilege to AI-assisted analysis; whether Heppner’s reasoning will carry into civil cases and Texas courts.