Delaware attorneys are joining colleagues nationwide in issuing urgent warnings to clients: conversations with artificial intelligence chatbots could end up as evidence in criminal prosecutions and civil lawsuits.
The cautionary advice has intensified following a significant federal court decision in New York earlier this year. A judge determined that a former financial company CEO facing securities fraud charges could not prevent prosecutors from accessing his conversations with AI chatbots.
Legal professionals are now alerting clients that discussions with popular AI platforms such as ChatGPT from OpenAI and Claude from Anthropic may be subject to court orders in both criminal investigations and civil litigation.
“We are telling our clients: You should proceed with caution here,” stated Alexandria Gutiérrez Swette, an attorney with the New York law firm Kobre & Kim.
While conversations between attorneys and their clients receive strong confidentiality protections under American law, AI chatbots do not qualify as lawyers. Legal professionals are now instructing clients on methods to better protect their communications with artificial intelligence tools.
More than twelve prominent U.S. law firms have distributed guidance through client emails and website postings, offering strategies to reduce the likelihood of AI conversations becoming courtroom evidence.
Some legal firms are incorporating these warnings directly into client contracts. Sher Tremonte, a New York-based firm, recently included language in a client agreement stating that sharing attorney advice or communications with a chatbot could eliminate attorney-client privilege protections.
The landmark case involved Bradley Heppner, former chairman of bankrupt financial services firm GWG Holdings and founder of alternative asset company Beneficent. Federal prosecutors charged Heppner last November with securities and wire fraud; he entered a not guilty plea.
Heppner had utilized Anthropic’s Claude chatbot to create reports about his case for sharing with his legal team. His attorneys argued that these AI conversations should remain confidential because they contained details from lawyers regarding his defense strategy.
Government prosecutors contended they had the right to obtain materials Heppner created using Claude, arguing that his defense attorneys weren’t directly involved and that attorney-client privilege doesn’t extend to chatbots.
Sharing information from a lawyer with any outside party can compromise the standard legal protections for attorney communications.
In February, Manhattan-based U.S. District Judge Jed Rakoff ordered Heppner to surrender 31 documents created using Anthropic’s Claude chatbot in connection with the case.
Judge Rakoff wrote that no attorney-client relationship exists “or could exist, between an AI user and a platform such as Claude.”
Neither Heppner’s legal team nor the Manhattan U.S. attorney’s office provided responses to requests for comment.
Courts are increasingly confronting the expanding use of artificial intelligence by both lawyers and self-represented individuals in legal proceedings, which has resulted in court filings containing fictional cases generated by AI systems.
Rakoff’s decision represented a crucial early examination in the AI era of fundamental legal protections covering attorney-client communications and materials prepared for litigation.
On the same date as Rakoff’s ruling, U.S. Magistrate Judge Anthony Patti in Michigan reached a different conclusion. He determined that a woman representing herself in an employment lawsuit against her former employer didn’t need to provide her ChatGPT conversations about her employment claims.
Judge Patti classified the woman’s AI discussions as part of her personal “work-product” for the case, rather than conversations with an individual whom her employer could seek to use in its defense.
ChatGPT and other generative AI programs “are tools, not persons,” Patti wrote in his decision.
The privacy and usage policies for both OpenAI and Anthropic indicate that the companies may share user data with third parties. Both companies also require users to consult qualified professionals before depending on their chatbots for legal guidance.
During a February hearing in Heppner’s case, Judge Rakoff observed that Claude “expressly provided that users have no expectation of privacy in their inputs.”
Neither OpenAI nor Anthropic representatives provided immediate responses to requests for comment.
Legal counsel recommendations have varied from advising clients to carefully choose their AI platforms to proposing specific language for chatbot interactions.
Los Angeles-based O’Melveny & Myers and other firms indicated in client advisories that “closed” AI systems designed for corporate use might offer stronger protections for legal communications, though they noted this remains largely unproven.
Several firms suggested that AI legal research is more likely to receive attorney-client privilege protection when conducted under lawyer supervision. If an attorney recommends using AI, individuals should indicate this in their chatbot prompt, according to New York-headquartered Debevoise & Plimpton in a website notice.
“I am doing this research at the direction of counsel for [X] litigation,” the firm recommended people include in their messages.
Details about AI usage are also becoming standard in law firm client contracts, based on a Reuters examination of contracts posted on a U.S. government website.
Sher Tremonte, which frequently represents white-collar criminal defendants, stated in a March contract: “Disclosure of privileged communications to a third-party AI platform may constitute a waiver of the attorney-client privilege.”
Justin Ellis from New York-headquartered MoloLamken and other legal professionals anticipate that additional court rulings will eventually establish clearer guidelines for when AI conversations can serve as evidence.
Until that clarity emerges, attorneys are emphasizing that a time-tested principle remains valid: avoid discussing your case with anyone except your lawyer — including artificial intelligence.
