The Chatbot in the Courtroom: What U.S. v. Heppner Means for Health Tech Founders Who Use AI to Think Through Legal Problems
Abstract
The February 10, 2026 bench ruling in U.S. v. Heppner (No. 25-cr-00503-JSR, S.D.N.Y.) is the first major federal decision holding that a client’s use of a consumer AI tool to prepare defense strategy documents waived attorney-client privilege. Judge Jed Rakoff’s reasoning rested on three pillars: (1) AI tools are not attorneys, (2) consumer ToS explicitly disclaim confidentiality and permit training on inputs and disclosure to gov’t authorities, and (3) pre-existing non-privileged documents don’t become privileged just by being sent to a lawyer after the fact. This matters disproportionately to health tech founders and executives who routinely use consumer AI to draft, summarize, or analyze content that touches on FDA regulatory submissions, HIPAA compliance memos, investor deal terms, IP licensing strategy, employment disputes, board communications, and reimbursement appeals. The ruling does not automatically doom enterprise AI use – courts are likely to treat tools that contractually prohibit training on inputs and maintain data confidentiality differently – but the line between “safe” and “not safe” is less obvious than most people think.
Key takeaways:
- Consumer AI (ChatGPT free tier, [Claude.ai](http://Claude.ai) free/Pro) = no privilege protection per Heppner
- Enterprise/API tiers with zero-data-retention = likely safer, not yet definitively tested in court
- AI note-takers on calls with counsel = almost certainly problematic
- Work product doctrine is a separate shield, and it also failed in Heppner
- The three-part test for privilege: attorney, confidentiality, legal-advice purpose – consumer AI fails all three
- Heppner had Quinn Emanuel and still lost – this isn’t just a DIY legal-research problem
- Health tech-specific exposure: HIPAA enforcement, FDA 483s, SEC/FTC investigations, M&A due diligence, cap table disputes
Table of Contents
What Actually Happened in Heppner
The Three Ways Privilege Dies
Work Product Doctrine: The Second Line of Defense That Also Collapsed
Why Health Tech Is More Exposed Than Most Sectors
The Enterprise Carve-Out: Real Safety or Just a Better Story to Tell Your Board
The AI Note-Taker Problem Nobody Is Talking About
Practical Protocol for Founders and Executives Right Now
What Actually Happened in Heppner
Before getting into implications, it helps to actually understand the facts, because a lot of the commentary floating around gets them a little wrong in ways that matter.
Bradley Heppner was the former CEO of Beneficient, an alternative finance company in Dallas. He got indicted in October 2025 on securities fraud, wire fraud, conspiracy, and making false statements to auditors in connection with an alleged scheme that prosecutors say cost GWG Holdings investors roughly $1 billion – GWG filed for bankruptcy, and prosecutors allege Heppner extracted over $150 million for himself before that happened, including $40 million to renovate his Dallas mansion. He was represented by Quinn Emanuel, which is as serious a white-collar defense firm as you can hire.
Here’s the key sequence. After Heppner knew he was a law enforcement target and had retained counsel, he used the consumer version of a commercial AI tool to run queries related to the government’s investigation. He fed information he’d learned from his Quinn Emanuel attorneys into the AI, generated 31 documents of prompts and AI responses, and then transmitted those documents to his lawyers. He apparently thought this would help him organize his thinking and prepare for the case. When the FBI executed a search warrant on his home, they seized devices containing those documents. His legal team asserted both attorney-client privilege and work product protection. The government moved to compel production. Judge Rakoff ruled from the bench on Feb. 10, 2026 that neither doctrine protected the documents, saying “I’m not seeing remotely any basis for any claim of attorney-client privilege.”
A few things worth flagging that people tend to glide past. First, Heppner wasn’t some naive founder who didn’t know he was in legal jeopardy – he was under active investigation, knew it, and had already engaged counsel. Second, he had actually received substantive legal strategy input from Quinn Emanuel attorneys and then incorporated that information into his AI queries. Third, his lawyers argued strenuously that the documents should be protected, and they didn’t win. If you’re imagining that “well, I was just using AI to prep some notes before calling my lawyer” puts you in a different category than Heppner, think that through more carefully.
The Three Ways Privilege Dies
