The Heppner Ruling: What It Means for Your Meeting Notes

In late 2025, federal agents executed a search warrant at Bradley Heppner's home and seized his devices. On those devices: 31 documents — private conversations he'd had with an AI chatbot while preparing his legal defense. He thought they were confidential. They weren't. The AI platform's privacy policy reserved the right to share user data with third parties — including government authorities. That one clause handed prosecutors his entire defense strategy.

On February 17, 2026, Judge Jed Rakoff of the Southern District of New York made it official, calling it "a question of first impression nationwide": anything shared with a consumer AI platform is not confidential. Heppner was convicted on all counts on May 7.

The case involved a chatbot. But the legal principle the court established reaches every cloud-based AI tool in a professional meeting room — and most people using those tools have no idea.

---

What happened

Bradley Heppner was the founder of Beneficient and former chairman of GWG Holdings, indicted in late October 2025 on charges of securities fraud, wire fraud, conspiracy, and falsification of records. His lawyers argued the 31 seized documents were privileged — that he had used Claude "in anticipation of a potential indictment" to help him "prepare reports that outlined defense strategy and what he might argue with respect to the facts and the law," and that those outputs had shaped counsel's strategy going forward.

The government disagreed. On February 10, Judge Rakoff ruled from the bench that none of it was privileged. His written opinion followed February 17. Three grounds. Each one worth understanding — because each one reaches further than just chatbots.

---

What the court decided — and why

First: Claude is not an attorney. Attorney-client privilege protects communications between a client and their lawyer. "Because Claude is not an attorney," Judge Rakoff wrote, "that alone disposes of Heppner's claim of privilege." Sharing AI output with your lawyer afterward does not make the AI a legal agent.

Second — and this is the one that matters for everyone else — the communications were not confidential. Anthropic's privacy policy "provides that Anthropic collects data on both users' 'inputs' and Claude's 'outputs,' that it uses such data to 'train' Claude, and that Anthropic reserves the right to disclose such data to a host of 'third parties,' including 'governmental regulatory authorities.'" Therefore, Judge Rakoff concluded, "Heppner could have had no 'reasonable expectation of confidentiality' in his communications with Claude." The moment his words touched a commercial server, confidentiality was gone.

Third: The purpose requirement was not met. Heppner created the documents on his own initiative, not at counsel's direction. Claude itself responded that it could not provide formal legal advice. No attorney direction. No legal advice. No privilege.

At trial, prosecutors used Heppner's AI prompts as active evidence against him. He was convicted on all counts.

---

Why this matters beyond chatbots

Here's what Heppner actually means for meeting rooms: it doesn't matter that he used a chatbot. The principle is about what happens when you send your data to a third-party server whose terms of service permit disclosure. That's not a chatbot problem. That's every cloud AI tool problem.

When you use Otter, Fireflies, Read AI, or any comparable cloud transcription service, the audio from your meeting is captured and transmitted to that company's servers. Every one of those services has a privacy policy that — like Anthropic's — reserves rights to collect, retain, and in some circumstances disclose that data.

Under Heppner's reasoning, any meeting transcribed through such a tool — a client strategy discussion, an investor conversation, a legal debrief — may no longer be considered confidential in the way participants assumed. The audio went somewhere. Someone else's terms govern it. A subpoena can reach it.

The Harvard Law Review's March 2026 analysis, written by Elizabeth X. Guo, noted the opinion "veers toward categorically excluding a client's use of generative AI from attorney-client privilege." Even critics of the ruling acknowledge the confidentiality risk is real. If the platform's terms permit disclosure, a court applying Heppner's reasoning may reach the same result.

---

The construction and engineering angle

For construction project managers, engineers, and consultants, the exposure is concrete. Client site meetings routinely involve proprietary project data, subcontractor negotiations, dispute discussions, and budget decisions that all parties treat as confidential. If those meetings are being transcribed by a cloud tool — even one marketed as "secure" — the audio and transcript exist on a third-party server under that company's terms.

In any project dispute or litigation, opposing counsel could subpoena the transcription service. Whether privilege protects those records depends on how the court applies the confidentiality analysis Heppner has now established. That's no longer a hypothetical question. It has an answer, and the answer isn't favorable.

McDermott Will & Emery partners Shawn C. Helms, Caitlin Howe, Jason D. Krieser, and Joseph B. Evans published guidance on February 19, 2026, warning that "entering privileged communications or legal strategy into a public AI tool can be analogous to disclosing them to an unprotected third party." A cloud meeting transcription tool is, structurally, a public AI model applied to your meeting audio.

---

What Heppner does and doesn't cover

The ruling does not say AI tools can never be used in privileged contexts. Judge Rakoff noted the outcome "could have been different" if counsel had directed Heppner to use Claude — in which case Claude "might arguably be said to have functioned in a manner akin to a highly trained professional who may act as a lawyer's agent." Enterprise AI deployments under formal confidentiality agreements and attorney supervision may be treated differently.

What the ruling establishes clearly: a consumer AI platform's privacy policy permitting data retention and third-party disclosure is fatal to a confidentiality claim. That's the test. Read the terms of any tool you use in a sensitive meeting. Most people don't. After Heppner, that's no longer a reasonable oversight.

---

The architecture that eliminates the risk

The Heppner confidentiality problem exists because data leaves the device. The moment audio is transmitted to a third-party server, the user no longer has exclusive control. That server's terms govern it. A subpoena can reach it. A breach can expose it.

Local-first transcription removes this at the architectural level. When audio is processed entirely on the device — in memory, without creating an audio file, without transmission to any server — there is no third-party platform involved. No privacy policy governing the data. Nothing to subpoena. The Heppner analysis has nothing to reach.

BarnOwl is built on this architecture. Audio is processed locally using on-device speech recognition. No audio file is created. No data leaves the machine. No cloud service is involved. The meeting stays in the room. In states requiring all-party notification — including California, Florida, Illinois, Pennsylvania, and Washington — users are recommended to inform all meeting participants that transcription is active before the meeting begins.

---

Frequently asked questions

What is the Heppner ruling?

United States v. Heppner, No. 25-cr-00503-JSR (S.D.N.Y. Feb. 17, 2026) is a federal court decision in which Judge Jed Rakoff ruled that 31 documents created by a criminal defendant using Anthropic's Claude chatbot were not protected by attorney-client privilege or the work product doctrine. The ruling addressed "a question of first impression nationwide." The defendant was convicted on all counts on May 7, 2026, and his AI-generated documents were used as evidence against him at trial.

Does the Heppner ruling apply to meeting transcription tools?

Not as a direct holding — the ruling addressed a consumer chatbot. But the court's confidentiality reasoning applies to any AI service that processes your data on third-party servers under terms permitting retention and disclosure. Cloud meeting transcription tools operate exactly this way. If the platform's privacy policy permits third-party data sharing, a court applying Heppner's reasoning may reach the same result.

Does this mean I cannot use AI tools in confidential meetings?

The architecture is what matters. Cloud-based tools that transmit audio to third-party servers carry the risk Heppner has now made concrete. Local-first tools that process audio entirely on-device — with no cloud transmission, no third-party retention — do not share that architecture and therefore do not share that risk. The question to ask of any tool: where does the audio go after I speak?

What should construction and engineering firms do now?

Review which AI meeting tools are in use across your organization. Read the privacy policies of those tools — specifically what they say about data retention, training use, and third-party disclosure. For meetings involving client confidentiality, NDAs, or legal strategy, evaluate whether a cloud tool is compatible with your confidentiality obligations. For future meetings with those constraints, a local-first tool is the safer choice.

What is the difference between a cloud transcription tool and a local-first tool?

A cloud transcription tool sends your meeting audio to a server operated by a third-party company, where it is processed and — depending on the terms of service — potentially stored, used for training, or disclosed in response to legal process. A local-first tool processes audio entirely on your own device, in memory, without transmitting it anywhere. Nothing is stored on a third-party server because nothing goes there in the first place.