- within Litigation and Mediation & Arbitration topic(s)
- with readers working within the Technology and Transport industries
Employees Are Turning to AI for Legal Advice and That Is a Problem
Today, many clients use artificial intelligence in their businesses to improve services, increase efficiency, and reduce costs. In many cases, employees use a combination of paid and free large language model artificial intelligence engines (“AI tools”), such as Claude, ChatGPT, and Gemini. Because of the ubiquity of these AI tools, employees may also be tempted to ask them legal questions that are better directed todirect legal questions to these tools rather than seeking guidance from in-house or outside counsel. A recent case highlights the dangers of doing so.
United States v. Heppner: When AI Meets Criminal Investigation
In United States v. Heppner (S.D.N.Y. Feb. 17, 2026), Judge Rakoff addressed whether a criminal defendant’s use of a public AI tool was protected by either the attorney-client privilege or the attorney work product doctrine. The case arose out of allegations that defendant Heppner defrauded investors by making false representations and causing a company of which he was an executive to enter into undisclosed self-dealing transactions.
How AI-Generated Legal Research Became Evidence
In 2025, after Mr. Heppner received a grand jury subpoena and it became clear that he was the target of an investigation, he prepared reports using Claude that outlined defense strategies and arguments he anticipated making with respect to both the facts and the law. In connection with Heppner’s arrest, the government executed a search warrant and seized documents containing the Claude-generated queries and reports.
Why Attorney-Client Privilege Does Not Protect AI Tool Use
In opposing disclosure of the queries and reports, Heppner argued that the materials were protected by the attorney-client privilege and/or the attorney work product doctrine. Judge Rakoff, in what may be the first published decision addressing whether a client’s use of an AI tool may qualify for protection under either doctrine, found that neither applied to Heppner’s use of Claude.
The Work Product Doctrine and the Limits of AI Confidentiality
In addition to observing that Claude is not an attorney, Judge Rakoff found that Claude’s privacy policy allowed user queries to be shared with third parties, including the government. As a result, Heppner lacked the requisite expectation of confidentiality necessary to invoke the attorney-client privilege. Judge Rakoff further held that the materials did not qualify as attorney work product because they were not prepared “at the behest of counsel” and did not reflect defense counsel’s legal strategy.
What the Heppner Decision Means for Client Use of AI Tools
What does this mean for counsel? First, counsel should educate clients that the use of public AI tools for self-help legal services is risky. Not only may public AI tools provide false or misleading answers, but the queries themselves may also become discoverable in litigation, much like other forms of electronic communications. Indeed, litigators may begin including AI-generated queries and responses in standard discovery requests for documents and electronically stored information.
Attorney Caution: AI Legal Research May Not Be Privileged
Second, the case serves as a reminder that attorneys should exercise caution before using public AI tools to conduct legal research. Although many attorneys already recognize the risks associated with AI-generated legal research — including hallucinations, fabricated citations, and inaccurate analysis — the queries and resulting output may also fall outside the protections of the attorney-client privilege and work product doctrine.
How Public AI Platforms Differ From Traditional Legal Research Tools
Attorneys have used computerized legal research tools for decades, but public AI tools differ in a critical respect: traditional legal research platforms are generally designed to maintain the confidentiality of user searches and results. Public AI platforms, by contrast, may retain, review, or disclose user inputs and outputs.
Limiting Heppner: When Private AI Tools May Still Qualify for Protection
Third, Heppner should not be read beyond its specific facts. The queries at issue were submitted to a public AI tool by a non-lawyer acting independently and not at the direction of counsel. Attorneys who use private AI tools — such as a secured internal AI instance or subscription-based legal AI platforms like Lexis AI — may still be able to invoke attorney-client privilege or work product protection. Nevertheless, as with all computerized research tools, any AI-generated results must be independently verified and carefully analyzed.
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.
[View Source]