In February 2026, the Southern District of New York issued a decision which will sure to be cited for years to come regarding the use of artificial intelligence (“AI”) and whether such information shared with AI platforms is discoverable in a legal proceeding. While everyone, regardless of industry, is seeking to understand and utilize AI before getting left behind, the court’s decision in United States v. Heppner is an important reminder that AI is not always the correct choice.
In the Heppner decision, the defendant, Bradley Heppner, was arrested in October 2025 after being indicted on a slew of securities and wire fraud claims stemming from him allegedly defrauding investors out of more than $150 million while making self-serving transactions between public entities for which he was an executive and privately held companies that he controlled. Upon his arrest, FBI agents seized Heppner’s electronic devices, which contained thirty-one documents that memorialized communications he had with the AI platform “Claude.” These Claude interactions all occurred between Heppner receiving a grand jury subpoena and his arrest.
Without his actual legal counsel suggesting doing so, Heppner prompted Claude to prepare reports that outlined a potential defense strategy and what he might argue with respect to the facts and the law that he anticipated would be part of the federal charges.
Heppner’s attorney asserted privilege over the information contained within his Claude account by claiming that (1) Heppner input information he learned from his counsel into Claude; (2) Heppner had created the AI documents for the purpose of speaking with counsel to obtain legal advice; and (3) Heppner subsequently shared the contents of the AI documents with his counsel. His counsel, however, conceded that he never instructed Heppner to run the Claude prompts.
It is well-settled that attorney-client privilege only attaches to “communications (1) between a client and his or her attorney (2) that are intended to be, and in fact were, kept confidential (3) for the purpose of obtaining or providing legal advice.” The Court found that Heppner’s AI documents certainly failed the first two of the above three-pronged test and likely the third, as well. The reasoning for each is briefly addressed below:
1. “between a client and his or her attorney”
Simply stated, Claude, or any other AI platform, is not an attorney. Despite the advancements of AI, an attorney-client relationship requires a licensed professional who owes a fiduciary duty and is subject to discipline for failure to abide by such duty. AI platforms unequivocally fail to meet that threshold.
2. “that are intended to be, and in fact were, kept confidential”
The information you input into Claude, which is owned by the company Anthropic, PBC, is not confidential. In fact, Claude’s written terms and conditions state that Claude users consent to Anthropic’s collection of data from users’ “inputs” and Claude’s “outputs” to better “train” Claude’s ability to provide responses. Moreover, Anthropic’s written policy also reserves the right to disclose such data to various third parties, including in connection with claims, disputes, or litigation. Therefore, Heppner voluntarily disclosed this information without the expectation of this information being kept confidential.
3. “for the purpose of obtaining or providing legal advice”
While the Court opined that this third prong likely also failed, it was rendered moot by the clear failure under either of the first two prongs. While Heppner’s counsel argued that Heppner input the information into Claude for the “express purpose of talking to counsel,” he also conceded that he never suggested that Heppner do so. Had his attorney made such a suggestion for the purpose of Heppner providing the Claude “output” to his attorney, an argument could be made that it was akin to a highly trained professional acting as the lawyer’s agent within the protection of the attorney-client privilege (for example, a construction industry claims consultant receiving information to calculate an equitable adjustment or create a time-impact analysis). Here, however, the facts do not support such an argument, because it was conceded that Heppner consulted Claude by his own volition.
Even if Heppner had intended to share the Claude “output” with his attorney, the Court bluntly stated that it is “black-letter law that non-privileged communications are not somehow alchemically changed into privileged ones upon being shared with counsel.”
The Court also analyzed potential privilege under the work product doctrine, which “provides qualified protection for materials prepared by or at the behest of counsel in anticipation of litigation or for trial.” This argument also failed because of the uncontested fact that the AI documents were not “prepared by or at the behest of counsel,” even if subsequently shared with his counsel.
It is indisputable that AI is changing the way we all live and work. It is a tool with immeasurable abilities that continue to expand exponentially. However, despite how intrusive it may feel – perhaps due to the ability to instantly consult Claude, ChatGPT, Grok, or any other AI platform from your phone – your AI “input” and “output” are not guaranteed privacy.