INTERVIEW ON THE PRICE OF BUSINESS SHOW, MEDIA PARTNER OF THIS SITE.

Recently Kevin Price, Host of the nationally syndicated Price of Business Show, interviewed Alexander Paykin.

The Alexander Paykin Commentaries

A federal court in Manhattan has now answered a question many lawyers have been asking quietly for months: if a client uses a public generative AI platform to think through legal issues, are those communications protected?

In United States v. Heppner, the answer was no. Judge Jed Rakoff held that a criminal defendant’s interactions with Claude were not protected by either the attorney-client privilege or the work product doctrine, where the platform was publicly available, the chats were not confidential, and the materials were not created at counsel’s direction.

For practitioners, the case is less about novelty than discipline. It is a reminder that privilege is easy to lose, especially when new technology is treated casually.

A first-impression ruling with immediate practical impact

The court described the issue as one of first impression: whether a user’s communications with a publicly available AI platform, in connection with a pending criminal investigation, are protected by privilege or work product. Judge Rakoff’s conclusion was straightforward: they are not, at least under these facts.

The defendant had used Claude after receiving a grand jury subpoena and after learning he was a target of the investigation. According to defense counsel, the defendant used the platform to prepare reports outlining defense strategy, anticipated charges, and factual and legal arguments. Counsel later claimed privilege over the documents and listed them on a privilege log. That is what brought the issue squarely before the court.

That procedural point matters. Heppner shows how AI use may surface in ordinary discovery mechanics. Once AI-generated or AI-assisted materials are identified, courts and adversaries will ask whether they are protected. Lawyers should expect that scrutiny going forward.

The court’s reasoning was old-school privilege law

One of the striking things about Heppner is that the opinion is not futuristic in tone. It is doctrinal and traditional.

The attorney-client privilege failed because Claude was not an attorney, there was no protected attorney-client relationship with the platform, and the communications were not confidential. The court relied heavily on the platform’s privacy terms, which allowed for data collection, model training, and disclosure in certain circumstances. On that record, there was no reasonable expectation that the communications would remain private.

The work product doctrine failed for a similarly conventional reason. Judge Rakoff emphasized that the doctrine protects materials prepared by or for counsel in anticipation of litigation, chiefly to protect the mental processes of lawyers. Here, the documents were created by the defendant on his own initiative, not at the request or direction of counsel, and they did not reveal counsel’s strategy as it existed when the materials were generated.

The court also rejected an argument that later disclosure to counsel changed the character of the materials. It did not. A client cannot retroactively cloak otherwise unprotected communications with privilege simply by sending them to a lawyer later.

The real lesson: AI use now belongs in client counseling

The biggest significance of Heppner may be practical rather than doctrinal.

Lawyers have long warned clients not to share privileged information with outsiders. Public AI systems now need to be included explicitly in that warning. Many clients do not see these tools as “third parties” in the traditional sense. They see them as software, search engines, assistants, or drafting aids. Heppner is a clear signal that courts may not view them that way.

That means counsel should be proactive.

At intake, lawyers should ask whether clients have used any AI tools in connection with the dispute, investigation, or transaction. During the representation, lawyers should repeat the warning, especially when the matter becomes contentious or document-heavy. Clients should be told not to upload legal memoranda, summaries of attorney advice, draft testimony, expert materials, or confidential facts into public AI tools.

Companies should also revisit internal training. This is not just a litigation issue. Compliance teams, HR professionals, finance staff, and operations personnel often use AI tools to summarize documents, draft communications, or analyze sensitive information. If those materials involve legal advice, regulated data, internal investigations, or trade secrets, the organization may be taking on serious risk.

Enterprise tools may be different — but only if they really are

The opinion focused on a publicly available, non-enterprise AI platform. That leaves room for future cases involving closed-universe or enterprise tools with negotiated confidentiality terms, limited retention, and stronger controls.

But Heppner should not be read as blessing all enterprise AI usage. It simply means the analysis may differ. Counsel still needs to understand the vendor’s terms, retention settings, training practices, access controls, and disclosure provisions. “Paid” does not automatically mean “private,” and “enterprise” does not automatically mean “privileged.”

The safest practice is still to separate public-AI experimentation from privileged legal work unless the tool has been specifically reviewed and approved for that purpose.

Looking ahead

Judge Rakoff ended the opinion by recognizing generative AI as a “new frontier” while reaffirming that longstanding privilege principles still apply. That closing observation is likely to shape how other courts approach similar disputes.

Heppner will likely become a frequently cited case in discovery disputes, internal investigations, and client counseling memoranda. It gives courts a roadmap, and it gives lawyers a warning.

The core message is simple: if a client shares legal analysis or sensitive matter-related information with a public AI system, a court may treat that act the same way it would treat disclosure to any other third party. Once that happens, privilege and work product arguments become much harder to sustain.

For lawyers and clients alike, that is the new reality.