Much like the internet, artificial intelligence (AI) is woven into the fabric of daily life. AI tools are used to streamline innocuous tasks such as scheduling appointments and taking notes during meetings, to perform complex functions such as synthesizing data and generating reports and to obtain recommendations on a wide range of topics via a user-friendly chat feature. While AI assistants can simplify complex concepts, users should be aware of the risks of using AI tools in connection with legal or confidential matters following recent court decisions.
U.S. v. Heppner: A nonlawyer’s interaction with AI tools may not be privileged
In U.S. v. Heppner, a judge in the U.S. District Court for the Southern District of New York recently ruled that a nonlawyer’s communications with a publicly available AI tool are not protected by the attorney-client privilege or the work-product doctrine.1
As background, Bradley Heppner used Claude, a conversational AI assistant developed by Anthropic, to establish a legal defense after he received a grand jury subpoena and engaged legal counsel, but before his indictment for securities fraud, wire fraud, conspiracy, false statements to auditors and falsifying corporate records. Heppner entered into Claude the information he learned from his criminal defense counsel, which generated reports containing defense strategies and arguments. While defense counsel did not direct Heppner to use Claude or any other AI tools, Heppner shared documents prepared with Claude with his defense counsel. Following Heppner’s arrest, the FBI seized documents and electronic devices from his home after executing a search warrant. The items seized included 31 documents related to communications with Claude, and prosecutors filed a motion for a ruling that the documents were not protected by the attorney-client privilege or the work-product doctrine.
The court ruled that the communications with Claude were not protected by the attorney-client privilege on multiple grounds. First, the communications were not between Heppner and his legal counsel, but were instead between Heppner and Claude, which is not an attorney. Second, Heppner’s communications with Claude were not confidential and there was no reasonable expectation of confidentiality because Anthropic’s privacy policy specifically states that it may collect and retain user data, use the data to train Claude, and disclose the data to third parties, including governmental regulatory authorities. Lastly, legal counsel did not direct Heppner to use Claude or any other AI tools; therefore, Heppner’s communications with Claude were not intended to obtain legal advice from Claude, especially since Claude expressly states that it cannot provide legal advice. The court went on to state that even if Heppner’s communications to Claude contained privileged information, the privilege was waived when the information was shared with Claude or any other third party.
The court further ruled that the communications with Claude were not protected by the work-product doctrine because Heppner acted alone when using Claude, without the direction of his legal counsel, and the communications with Claude did not reflect defense counsel’s strategy.
Several emerging court decisions on AI tools and privilege
With many variables to weigh, court review of AI interactions and privilege will likely require a fact-specific analysis that may lead to narrow decisions about whether AI communications are discoverable or protected by the attorney-client privilege or the work-product doctrine.
Courts will need to address whether there is a meaningful difference between using a proprietary AI tool instead of a publicly available AI assistant. In Heppner, a publicly available version of Claude was used rather than an enterprise version.
The AI tool’s privacy policy will also likely factor into how courts determine if communications with an AI assistant are confidential and whether the user has a reasonable expectation of confidentiality. This may depend on whether the AI assistant collects and retains data, if it may disclose data to third parties and if the data is used to train the tool to improve its output.
Courts may be more inclined to rule that a user’s conversations with AI assistants are protected by the attorney-client privilege or the work-product doctrine if the AI tool was used at the direction of counsel and if the communications reflect legal counsel’s mental impressions and strategy.
Future cases will determine how Heppner will be distinguished, and one court has already reached a contrary conclusion. In Warner v. Gilbarco, Inc., a magistrate judge in the U.S. District Court for the Eastern District of Michigan recently ruled that documents and information related to a self-represented party’s use of AI tools are not discoverable and are protected work product.2 Warner is a civil employment discrimination case brought by a pro se plaintiff who used ChatGPT, a conversational AI assistant developed by OpenAI, to prepare her case. The court ruled that a pro se litigant’s internal analysis and mental impressions formed using ChatGPT are protected work product and are not discoverable.
Special considerations when integrating AI tools into business operations
After Heppner, companies should review their practices regarding the use of AI tools in connection with legal or confidential matters, especially when those matters involve liability concerns or potential or actual third-party claims or litigation. AI tools use and create an extraordinary amount of data composed of chat prompts and responses, uploaded files and generated and recorded text, audio and visual content. All this information may be evidence sought via subpoenas, discovery and search warrants.
As the use of AI assistants becomes commonplace, organizations are realizing that integrating AI into business operations has outpaced oversight of AI, and that they need to align their AI policies with how the technology is actually used. This is especially true as employees are embracing AI and using AI assistants in performing their job duties — even if the company has not explicitly permitted the use of any specific AI tools. Businesses should identify what types of AI tools are used in their operations and determine the parameters of their use, and whether the tool may access data that is confidential, proprietary or owned by a third party, such as a customer or client. It’s essential for companies to understand the AI tool’s privacy policy and data practices.
Organizations should be aware of situations in which using AI assistants may compromise confidentiality. Companies should provide employees with guidance on the appropriate use of AI tools, including AI notetakers that may transcribe meetings and generate summaries of topics discussed. If employees are authorized to use AI notetakers in meetings, they should be advised to disable the tool during confidential discussions or when speaking with legal counsel to prevent inadvertent disclosure, waiver of privilege or unauthorized access to attorney-client privileged communications that would otherwise be transcribed, collected and disseminated by an AI notetaker.
Organizations using AI assistants in their business operations should be prepared to defend their reliance on AI output, including its impact and influence on company decisions. When adopting and integrating any new technology or practice into business operations, it’s essential that companies weigh the benefits of efficiency with the risks, particularly the risk of disclosure.
HUB’s professional liability specialists can help your organization position its use of AI within a sound risk management strategy. Connect a HUB ProEx Specialist to review your AI policies and identify opportunities to strengthen your approach. View more articles in HUB’s ProEx Advocate Articles & Insights Directory.
NOTICE OF DISCLAIMER
Neither HUB International Limited nor any of its affiliated companies is a law firm and therefore cannot provide legal advice. The information herein is provided for general information only and is not intended to constitute legal advice as to an organization’s or individual’s specific circumstances. It is based on HUB International’s understanding of the law as it exists on the date of this publication. Subsequent developments may render this information outdated or incorrect, and HUB International has no obligation to update it. You should consult an attorney or other legal professional regarding the application of the general information provided here to your organization’s specific situation and particular needs.
1 Harvard Law Review, “United States v. Heppner (Case 1:25-cr-00503-JSR),” March 23, 2026.
2 Govinfo, “24-12333 – Warner v. Gilbarco, Inc . et al,” December 4, 2025.