By Michael R. Huttenlocher, Lewis Zirogiannis, P.C., Yoav Gaffney, Thomas F. Li and Abigail Blaker

Companies, executives and investors increasingly rely on AI tools to draft correspondence, summarize key materials, support decision-making and prepare for discussions with counsel. While using AI is not inherently problematic, the scope of attorney-client privilege and work product protections for AI-related exchanges remains unsettled. Two recent decisions—United States v. Heppner (“Heppner”) from the Southern District of New York and Warner v. Gilbarco, Inc. (“Gilbarco”), from the Eastern District of Michigan—illustrate how these protections may apply depending on different circumstances. Although these rulings reached different outcomes, they underscore the same practical lesson: using AI tools in legal workflows may pose risks unless specific safeguards are in place.

Background

Heppner

After learning he was the target of a federal criminal investigation, Bradley Heppner used Anthropic’s AI platform Claude to prepare a defense for his potential indictment and litigation. Without direction from his attorneys, Heppner input information his attorneys had shared with him into Claude and engaged in several exchanges with the tool, generating approximately 31 documents outlining potential defense strategies and Heppner’s view of the facts and law. He then shared Claude’s outputs with his defense counsel. When the FBI seized the documents during Heppner’s arrest, his counsel asserted attorney-client privilege and work product protection over the Claude exchanges. The Government moved for a ruling that neither protection applied.

Gilbarco

In an employment discrimination case, pro se plaintiff Sohyon Warner used OpenAI’s ChatGPT to draft filings and analyze legal questions. The defendants moved to compel production of all AI-related materials, arguing that the plaintiff waived work product protection by disclosing the materials to a third-party platform.

The Court’s Holdings

Heppner: No Protection

On February 17, 2026, Judge Rakoff held that Heppner’s exchanges with Claude—a publicly available AI platform that permits third-party disclosure of user inputs—were not protected by the attorney-client privilege or the work product doctrine.

The Court rejected Heppner’s attorney-client privilege claim on multiple grounds. First, Claude is not an attorney and discussions of legal issues between non-attorneys receive no protection. Second, Heppner’s communications with Claude were not confidential. The Court emphasized that Anthropic’s privacy policy explicitly permits the company to collect user inputs and outputs, use that data for AI training and disclose it to third parties, including government authorities, thereby removing any reasonable expectation of confidentiality, even with respect to information Heppner had originally received from counsel. Third, Heppner did not communicate with Claude to obtain legal advice; although Heppner later forwarded the documents to his counsel, that did not retroactively protect them from disclosure.

The Court also rejected Heppner’s work product claim. The work product doctrine protects materials prepared by or at the direction of counsel in anticipation of litigation. Here, Heppner prepared the AI documents of his own volition, and although Heppner had input information he received from counsel, the documents did not reflect counsel’s strategy at the time Heppner generated them.

Gilbarco: Work Product Protection Applies

On February 10, 2026, Magistrate Judge Patti held that Warner’s exchanges with ChatGPT were protected by the work product doctrine.

The Court explained that Warner, who was self-represented, was using ChatGPT to prepare legal filings and that her exchanges reflected her legal strategy. Moreover, disclosure to ChatGPT had not waived the protection: the court explained that work product protection “is waived only when documents are used in a manner contrary to the doctrine’s purpose,” specifically by disclosure “to an adversary or in a way likely to get in an adversary’s hand.” According to the Court, generative AI tools “are tools, not persons, even if they may have administrators somewhere in the background,” and regardless, Warner’s disclosure to ChatGPT was not likely to result in disclosure to Gilbarco in ordinary circumstances.

Reconciling the Two Decisions

Although the two cases reach different outcomes, they are not necessarily inconsistent. The pro se footing in Gilbarco meaningfully shaped the analysis: because the plaintiff represented herself, her AI exchanges reflected her own mental processes and case strategy. By contrast, the Heppner court emphasized that Heppner’s exchanges did not reflect his counsel’s strategy and the AI platform Heppner used permitted disclosure to the government—Heppner’s actual adversary.

Implications for Companies, Executives, and Investors

Companies, executives, and investors increasingly use AI to support core workflows, including legal: negotiating agreements, managing compliance matters, assessing potential risks and liabilities, responding to regulatory inquiries and preparing for litigation.

These decisions carry several key lessons. First, using generative AI in legal workflows should be limited to platforms with strong confidentiality protections, typically enterprise deployments. The ruling in Heppner is limited to AI platforms whose terms permit data collection and third-party disclosure, including government authorities. Second, AI-generated material intended for use in litigation should be prepared at counsel’s direction. Heppner’s decision to act independently was fatal to his attorney-client privilege and work product claims; represented parties should ensure AI use occurs within the attorney-client relationship. Third, the law continues to evolve. Heppner attracted significant attention, but Gilbarco shows courts may take different approaches to different sets of facts. Accordingly, businesses should monitor developments and consult counsel when developing AI policies—and, while the law develops, we recommend adopting a cautious approach. Consider at least the following steps:

Practical Steps for Companies, Executives and Investors for Using AI in Legal Workflows

1. Keep privileged materials out of AI platforms. Do not input legal documents or memos, attorney emails, or counsel’s advice into non-enterprise AI platforms.
2. Use enterprise AI with meaningful protections. Review privacy policies for data sharing, training use and disclosure to authorities, and require contractual commitments prohibiting training on user data and third-party disclosure. Merely toggling off AI training options is helpful but may not be adequate.
3. Use consumer-facing AI tools sparingly. Treat non-enterprise AI as a third party. Anything you input and any output you receive may be produced in litigation or in response to regulatory inquiries.
4. Ban AI notetakers from privileged meetings. Do not use AI transcription on calls with counsel or internal discussions of legal advice.
5. Involve counsel early. When disputes, investigations, or regulatory matters arise, use of AI should only be at counsel’s direction—not independently undertaken—to preserve attorney-client privilege and work product protection.
6. Train broadly. AI policies should reach board members, executives and employees, not just in-house counsel.

Conclusion

Generative AI is transforming how companies and investors operate. Heppner and Gilbarco demonstrate that longstanding legal doctrines continue to apply to new technologies and that privilege and work product determinations will often turn on fact-specific considerations. These decisions do not mean businesses should stop using AI. But businesses should be smart and intentional in how they use AI, with special attention paid when legal issues are involved.