As artificial intelligence tools become integrated into daily workflows, individuals and legal clients face unintended consequences regarding their confidential communications. Utilizing popular large language models such as ChatGPT, Claude, and Gemini can introduce significant risks, potentially stripping away the legal protections historically afforded by the attorney-client privilege and work product doctrine. Recent judicial rulings across United States courts highlight how sharing sensitive information with public-facing AI platforms can jeopardize legal confidentiality. ## The Fragility of Attorney-Client Privilege with Public AI Attorney-client privilege protects private communications between a client and their attorney intended for obtaining legal advice, ensuring clients can speak candidly without fear of disclosure to adverse parties. However, federal courts are increasingly scrutinizing what happens when those communications cross over into public AI tools. In *United States v. Heppner*, a federal district court judge in New York addressed a scenario where a client shared communications received from legal counsel with a public AI platform. The court determined that this action compromised confidentiality, effectively waiving the privilege for those specific communications. The presiding judge rejected arguments that the public AI functioned as an agent of the legal counsel—a status that can sometimes preserve privilege. The court based its rejection on two key points: * The AI platform is not a human being. * The attorney never explicitly directed the client to share the communication with the AI. Consequently, information fed into public AI models can become discoverable and subject to disclosure to opposing parties in ongoing legal matters. This risk also extends to automated AI notetakers integrated into popular video conferencing platforms. If allowed to run automatically during calls between attorneys and clients, these tools may capture and record sensitive legal discussions, undermining confidentiality. ## Nuances of the Work Product Doctrine and Pro Se Litigants While traditional attorney-client privilege faces strict limitations when exposed to public AI, rulings regarding the Work Product Doctrine—which protects legal writings, mental impressions, conclusions, and legal theories prepared in anticipation of litigation—have varied depending on representation. * Warner v. Gilbarco: In a Michigan federal district court, a pro se litigant representing herself in an employment dispute interacted with a public AI. The judge ruled that her interactions were not discoverable under the Work Product Doctrine, reasoning that the AI acted as a "tool" rather than a person, meaning she had not disclosed information to an adversary. * Following Opinions: Federal district courts in Colorado and Arkansas subsequently referenced and followed the *Gilbarco* reasoning, concluding that pro se litigants utilizing AI tools did not waive work product protections regarding litigation strategies. * Shealy v. Seaside Invs., LLC: Conversely, the Superior Court of Massachusetts reached a different conclusion when a represented party was involved. The court ruled that work product privilege did not protect AI-generated documents created by a party's romantic partner using ChatGPT, because the content was not prepared by or at the request of legal counsel, aligning the case more closely with the principles established in *Heppner*. ## Best Practices for Safeguarding Legal Confidentiality As AI tools continue to weave into everyday routines, legal experts emphasize that convenience often carries hidden costs. Individuals anticipating litigation or seeking to maintain strict confidentiality should adopt specific precautionary measures: * Treat public-facing AI platforms as third parties, assuming that any text entered may not remain confidential. * Avoid pasting sensitive facts, private documents, or communications from counsel into AI programs unless specifically approved by an attorney. * Monitor video calls to ensure unsecure AI notetakers are not automatically recording, and remove them before discussing legal advice.