Every attorney whose clients use ChatGPT, Claude, or other consumer public generative AI tools should pay close attention to what happened in a Manhattan federal courtroom this February.

On February 10, 2026, Judge Jed S. Rakoff of the U.S. District Court for the Southern District of New York decided that documents a defendant made with a consumer AI tool and later gave to his lawyer were not protected by attorney-client privilege or the work product doctrine. The defendant, Bradley Heppner, was indicted for fraud. While awaiting trial, he used an AI platform to prepare reports on his defense strategy and potential legal arguments, then included those 31 documents in his privilege log.

The court found no basis for attorney-client privilege because the communications were with an AI tool, not a lawyer, and were only later sent to counsel. Critically, Judge Rakoff noted that the platform’s written privacy policy expressly reserved the right to disclose user data to third parties, including governmental regulatory authorities, and to collect user inputs to train the model. On that basis, Heppner could have had no reasonable expectation of confidentiality. The privilege log was rejected, and the documents were deemed discoverable.

What Your Client Does Before You Get Involved Matters

Most attorneys start thinking about privilege as soon as they are hired. But your client’s involvement with their case began before they ever contacted you.

If non-lawyers use generative AI to research a new legal claim or draft messages before hiring a lawyer, those AI chat histories could be discovered and possibly used as evidence against them in court.

This is the gap the Heppner case revealed. Heppner said he wanted to organize his thoughts and share them with his lawyer. But the court found he created the documents without his lawyer’s guidance and used a third-party platform that did not promise confidentiality. The prosecution also emphasized a point that trips up many clients: sharing non-privileged documents with an attorney does not make them privileged.

Judge Rakoff acknowledged that things might have changed if Heppner’s lawyer had directed him to use the AI tool, which could have made it an agent of the lawyer. In other words, if an attorney specifically directs a client to use an AI platform for case-related tasks, within the context of the attorney-client relationship and with clear confidentiality instructions, those communications may be more likely to stay protected by privilege. To help preserve privilege, attorneys should guide clients in how and when to use AI tools, ensure any usage is documented as part of providing legal advice, and use platforms that promise confidentiality. This is a practical opportunity to educate clients from the outset and establish privilege-protective practices around emerging technology.

The ESI Angle Attorneys Need to Flag

The Heppner decision directly affects how you approach electronic discovery in your cases.

AI-generated documents are now clearly a type of potentially discoverable ESI. Clients who used ChatGPT, Claude, Copilot, Perplexity, Gemini, or similar generative AI tools to research their case, draft timelines, or consider their risks before contacting you may have created documents that opposing counsel can request, and courts can require.

Consider these initial steps for identifying, collecting, and preserving AI-generated ESI:

• Ask clients directly about any use of public or private AI tools related to the matter, including both research and draft documents.
• Request access to relevant AI chat histories, download logs, or transcripts. Many AI tools provide options to export conversations or history.
• Preserve this data promptly, as some platforms only retain history for limited periods or allow users to delete records.
• Document each step of collection, noting the platform, dates of use, and whether any information was shared externally.
• Coordinate with your IT team or an ESI specialist if needed, as accessing and maintaining the integrity of AI-generated data may require unique technical steps.

The use of AI in legal practice has likewise changed how lawyers should approach client intake. Updating your intake process to ask specifically about AI tool use connected to the matter is a concrete, practical step attorneys can take now. Add these questions to your intake process:

• Has your client used any AI tool to research the facts of this matter, draft communications, or think through their situation?
• Were any of those outputs shared with third parties, or created on platforms with terms of service that disclaim confidentiality?
• Does your client’s organization have an AI use policy, and was the tool used inside or outside of it?

Some legal experts have raised whether a broad reading of Heppner could affect privilege whenever a non-human platform is used, such as Gmail, Westlaw, Lexis, or eDiscovery tools. That broader reading has been contested, and the law is still developing. The most widely accepted and cautious approach today is to treat information as unprotected when it is shared with or generated by third-party tools that do not guarantee confidentiality.

Most authorities currently view tools that function as conduits for communication between a client and their attorney, or that have strong privacy protections and attorney supervision, as lower risk from a privilege standpoint. Enterprise-level tools may be acceptable so long as they comply with data privacy and compliance policies. In the face of legal uncertainty, attorneys are increasingly advised to avoid using public-facing or consumer AI tools for privileged work unless confidentiality can be clearly assured and documented.

The main idea is simple: if a client shared sensitive information with a platform that does not protect confidentiality, those communications could show up in discovery.

Getting Your Arms Around It Before the Case Gets Away From You

The real challenge is that most clients will not mention this on their own. They used an AI tool because it seemed useful and private, and they did not realize they were creating documents that could matter in discovery.

Getting ahead of this requires two things. First, client communication at intake that specifically asks about AI tool use in connection with the matter. Second, a discovery process that accounts for AI-generated ESI as a potential data source, not just a novelty.

Working with counsel and client to identify where AI data resides, what platforms your client uses, what data those platforms retain, and how to preserve it appropriately should be part of your defensible case preparation.