Thinking about using ChatGPT, Claude, or Perplexity to collect opinions and email them to your attorney? Don’t assume your chat will remain confidential.
A federal judge ruled Tuesday that prosecutors can access Claude chat records created by Brad Heppner, the founder of a financial startup who is accused of defrauding companies out of $150 million.
The chats took place after Heppner received a subpoena, hired a lawyer and learned he was a target of prosecutors, his lawyer said in court.
Mr. Heppner, who helped found the financial firm Beneficient, was arrested last year and charged with wire fraud and securities fraud for actions that allegedly led to the collapse of GWG Holdings.
Prosecutors said investigators seized “dozens of electronic devices” when they arrested Mr. Hepner at his Dallas mansion, and Mr. Hepner’s lawyers argued that 31 chats with Anthropic’s Claude Bott on those devices were privileged.
“Mr. Heppner used AI tools to produce a report outlining a defense strategy and outlining the factual and legal claims we anticipated the government might make,” the lawyers said.
“His purpose in producing these reports was to share them with us in order to discuss defense strategy with us.”
Heppner had a privileged conversation with his attorney, according to the hearing transcript, but “disclosed it to a third party, an AI, with an express provision that the submission was not confidential,” Judge Jed Rakoff said.
The government pointed out that Claude’s privacy policy clearly states that the contents of his chats may be made public. Prosecutors also said the chats were not protected by the “work product privilege,” which protects materials created at the direction of a lawyer, because Hepner’s lawyer did not ask him to use Claude.
Lawyers are upset about this decision.
“My gut reaction is that this decision is directionally correct,” attorney Moishe Peltz, whose posts about the decision bounced around X, told Business Insider. “There’s a lot of material that people are putting into AI that should be privileged.”
Another wrote that the proliferation of chatbots, where people enter sensitive legal information, is creating a “discovery nightmare.”
Employment lawyer Noah Bunzle told Business Insider that people may find it “somewhat shocking” that sharing legal secrets with a chatbot could result in legal secrets being lost.
This case is not the first time that executives’ use of chatbots has been the subject of legal controversy.
In November, PC Gamer reported on a dispute involving the acquisition of a video game company. There, court records say company executives used ChatGPT to avoid earnout payments.
And after the New York Times sued OpenAI for allegedly infringing the copyright of news articles, a judge required OpenAI to keep millions of chat logs for potential copyright infringement review.
Bunzl said he has noticed that attorneys are increasingly requesting adversary AI chats in civil discovery. It’s “a whole other world of discoverable information,” he said.
Still, lawyers at law firm Debevoise & Plimpton, who analyzed the Heppner decision, said they were aware for the first time that someone’s use of an AI tool could result in a “loss of privilege” over privileged content. They said courts could take a different view of companies’ use of specialized AI tools.
Arlo Devlin Brown, a white-collar defense attorney, told Business Insider he believes AI models have the potential to improve attorney-client information. However, given the ambiguity of the law, people should be cautious.
“Until the law is clearer, lawyers should warn their clients that entering privileged information into AI tools risks exposure in litigation,” he said in an email.
Representatives for the U.S. Attorney’s Office for the Southern District of New York, Anthropic, and Mr. Heppner’s attorney did not immediately respond to requests for comment.
