Table of Contents
ToggleWhen a business owner accused of fraud inputs details of his case into an AI chatbot in an attempt to prepare for discussions with a lawyer, he may assume that those conversations are protected by law.
But in February of this year, Manhattan federal judge Jed Rakoff ruled that former GWG Holdings chairman Bradley Heppner must hand over 31 files he generated using Anthropic Claude to federal prosecutors, breaking the boundaries of lawyer privilege in the age of AI chatbots.
First ruling: AI chatbots are not a lawyer's secret
According to Reuters , Bradley Heppner, former chairman of the bankrupt financial services firm GWG Holdings and founder of the private alternative asset management company Beneficent, was indicted by federal prosecutors in November 2025 on charges of securities fraud and wire fraud, which he has denied.
During the trial, Heppner used Claude, an AI assistant from Anthropic, to write several reports on the case, which he planned to share with his defense attorney. His legal team argued that these AI conversations should be protected under attorney-client privilege because they contained details of the defense strategies provided by the attorney.
Federal prosecutors countered that the files were generated by Heppner itself, and that defense attorneys were not directly involved; the AI platform does not have the status of a lawyer, and legal privileges should not be applied.
Judge Rakoff's ruling supported the prosecutor's position, stating explicitly in his written opinion that there was "no, and could not be, an attorney-client relationship" between AI users and the Claude platform.
This case is regarded by legal scholars as a landmark ruling that marks the first direct judicial challenge to the protection of lawyer-client communications in the era of AI chatbots.
The legal profession is shaken: lawyers must carefully plan their wording before using AI in litigation.
Following the ruling, major U.S. law firms quickly updated their client warnings and contract terms.
The New York law firm Debevoise & Plimpton suggests that to ensure AI-generated profiles remain protected, users must explicitly state in the prompt: "I am conducting this research under the guidance of an attorney for the XX litigation."
The law firm Sher Tremonte added a warning to its new contract: "Disclosing privileged communications to third-party AI platforms may constitute a waiver of attorney-client privilege."
On the same day, on the same issue, a Michigan judge reached a contrary conclusion.
However, on the same day that Rakoff made his ruling, Michigan U.S. District Judge Anthony Patti reached a completely different conclusion on the same issue in another employment discrimination lawsuit.
The case involved a woman's conversations with OpenAI ChatGPT. Judge Patti positioned AI as a "tool, not a person," and considered these conversations as the individual's work, thus granting them protection.
The two opposing rulings reflect the significant divergence within the US judicial system regarding the legal status of AI communications. Legal scholars point out that similar conflicts may ultimately require appellate courts or even the Supreme Court to establish unified precedents.
OpenAI and Anthropic: Users should not have had privacy expectations in the first place.
In fact, the privacy policies of the two major AI companies have already clearly stated this. According to the terms of service of OpenAI and Anthropic, users should not have "privacy expectations" of conversations with AI platforms, and should consult qualified legal professionals before relying on AI tools to handle legal matters.
However, the reality is that, faced with exorbitant legal fees and complex legal procedures, more and more people are accustomed to using AI to "warm up" first: organizing their thoughts and clarifying the overall framework before heading to a law firm. The legal risks of this approach have never been so clearly examined by the judiciary before.
