Generative AI has quietly slipped into the way we work, doing everything from drafting emails and summarizing documents to brainstorming business strategy. But a recent federal ruling out of the Southern District of New York is a wake-up call: the conversations you have with an AI assistant may not be as private as you think, especially the event of litigation.

The Case in Brief

In United States v. Heppner, Judge Jed Rakoff confronted what the Harvard Law Review called "a question of first impression nationwide": whether a criminal defendant's written exchanges with Anthropic's Claude AI were protected by attorney-client privilege or the work product doctrine.

The defendant, Bradley Heppner, was indicted on fraud and false-statement charges in late 2025. After being notified he was a target of investigation, he used Claude — on his own initiative — to outline potential defense strategies and arguments. He then shared those AI-generated outputs with his attorneys, who incorporated them into their planning.

Judge Rakoff ruled the documents were not privileged. His reasoning rested on three points:

  1. Claude is not an attorney. The communications weren't between client and lawyer.
  2. No reasonable expectation of confidentiality. Heppner had shared information with a third-party platform whose privacy policy allows data to be used for training and disclosed to regulators or in litigation.
  3. No intent to obtain legal advice. Heppner wasn't directed by counsel to use Claude, and Claude itself disclaims giving legal advice.

Why It Matters Beyond the Courtroom

The Harvard Law Review commentary argues the opinion may sweep too broadly. Despite vendors’ ability to access data, courts routinely treat Gmail, Google Docs, Slack, and iCloud as privileged intermediaries. Why should AI be categorically different, especially when a user has opted out of data collection and is using the tool to prepare questions for counsel?

The deeper issue is that Heppner effectively prevents a client's self-directed AI use from being privileged unless the attorney explicitly directs it. That's a rule with real consequences for how professionals, executives, and everyday users engage with AI tools.

Practical Takeaways for Business Leaders

Whether or not future courts adopt a more flexible standard, here's what the ruling means right now:

  • Assume AI chats are discoverable. Treat anything you type into a public AI tool as potentially producible in litigation or investigation. This is not necessarily the case with paid enterprise-tier licenses with stricter data protection policies.
  • Loop in counsel early. If you're using AI to think through legal, regulatory, or compliance questions, do so only at your attorney's direction and document that direction.
  • Check your privacy settings. Many AI platforms, including Claude, let users opt out of training data collection. Review enterprise-tier agreements for stronger confidentiality protections.
  • Separate sensitive work. Keep high-stakes deliberation on platforms with clear confidentiality terms as opposed to consumer-tier chatbots.
  • Train your team. Employees experimenting with AI on company matters need guardrails, just as they did when Gmail and Dropbox first arrived.

The Bigger Picture

Heppner is the first word on AI and privilege — not the last. As the Harvard Law Review notes, AI "has become part of everyday life," and courts will continue charting new territory at the intersection of AI, privacy, and legal doctrine. The organizations that adapt thoughtfully now will be the ones best positioned when the rules catch up to the technology.