AI vs. Attorney-Client Privilege: Lessons from United States v. Heppner

Client Alert

Intellectual Property Update

Colin Raufer - RauferPatentLaw.ai

Monday, April 13, 2026

Topic: Case Spotlight

Important Disclaimer:

This client alert is provided for informational purposes only and does not constitute legal advice. It is not a substitute for consulting qualified legal counsel regarding your specific circumstances. Laws and interpretations evolve rapidly, and outcomes depend heavily on particular facts.

Background

In United States v. Heppner, No. 25-cr-00503-JSR (S.D.N.Y.), a federal court addressed a question of first impression: whether documents and communications generated by a criminal defendant using a publicly available generative AI platform (Anthropic’s Claude) qualify for attorney-client privilege or work product protection.

Bradley Heppner, a corporate executive facing securities fraud charges, independently used the AI tool after receiving a grand jury subpoena. He inputted facts and strategy ideas to outline defenses and prepare arguments, then later shared the outputs with his attorneys. During a search, the government seized the materials. On February 10, 2026, Judge Rakoff ruled from the bench (with a written opinion issued February 17) that the AI-generated documents were not protected. The court held they failed core elements of privilege: no attorney-client relationship existed because Claude is not a lawyer; there was no reasonable expectation of confidentiality due to the platform’s public terms and training practices; and the materials were not prepared at counsel’s direction for obtaining legal advice.

Why This Matters

This ruling highlights how traditional privilege doctrines apply strictly in the AI era. Corporate executives and in-house teams frequently use generative AI for quick analysis, brainstorming, or drafting in complex legal, compliance, or IP-related matters. When those tools are consumer-grade and used unilaterally, sensitive inputs and outputs can lose protection and become discoverable in investigations, litigation, or regulatory probes. The decision underscores that simply sharing AI outputs with counsel afterward does not create privilege retroactively.

Two Biggest Risks for Corporate Executives

1 - Loss of Attorney-Client Privilege and Work Product Protection

Executives who input confidential company information, legal strategies, or analysis derived from counsel discussions into public AI tools risk having those materials treated as non-privileged. In Heppner, the court emphasized that the AI is a third party, not an attorney, and unilateral use (even if intended to facilitate later talks with counsel) fails to meet privilege requirements. This can expose internal deliberations, admissions, or defense outlines to prosecutors, regulators, or adversaries, potentially weakening the company’s or executive’s position in litigation or investigations.

2 - Unintended Waiver Through Lack of Confidentiality and Third-Party Disclosure

Public AI platforms often lack robust safeguards—user prompts may be reviewed, used for model training, or disclosed under privacy policies. In Heppner, the absence of a reasonable expectation of confidentiality (due to Claude’s terms) was fatal. For executives handling sensitive IP portfolios, trade secrets, patent strategies, or compliance matters, feeding in such data can result in waiver, loss of exclusivity rights, or broader data security and regulatory exposure beyond mere privilege issues.

Two Practical Mitigation Strategies

1 - Route All AI Use Through Counsel’s Direction and Supervision

Involve outside or in-house counsel explicitly before using AI for any legal analysis, strategy outlining, or document preparation. Document the counsel’s instructions and treat the AI as a potential agent under doctrines like Kovel (which can extend privilege to necessary third-party assistants). Route all inputs/outputs through counsel and use only tools provisioned or approved by the legal team. This helps satisfy the requirements that materials be prepared “by or at the behest of counsel” in anticipation of legal advice.

2 - Adopt Enterprise-Grade AI Tools with Strong Confidentiality Protections

Avoid consumer/public AI platforms for privileged or sensitive work. Instead, deploy enterprise versions with contractual no-training-on-user-data policies, data isolation, and confidentiality commitments. Establish clear internal AI governance policies requiring privilege reviews before any inputs, limit use in regulated or IP-sensitive contexts, and provide regular training for executives. Consider counsel-controlled environments to minimize control and conflict issues.

Educational Takeaways

1. Traditional privilege rules still apply strictly—AI does not automatically inherit protections.
2. Counsel involvement is key; post-hoc sharing is insufficient.
3. Have you considered auditing your team’s current AI usage habits for any legal or strategic tasks?

Have You Considered?

Have you considered updating your company’s AI usage policy to require counsel pre-approval for any prompts involving confidential, legal, or IP-related content?

We recommend reviewing your organization’s AI governance policies, internal training programs, and privilege protocols in light of this development. For a deeper dive, a privileged audit of your specific AI practices, or assistance implementing these strategies, please reach out to RauferPatentLaw.ai IP or litigation counsel.

This alert is for general educational purposes only. This area of law continues to develop quickly.

Prepared by: Colin Raufer