United States v. Heppner: what the AI privilege ruling means for legal tech and access to justice

What happened in United States v. Heppner?

The facts of United States v. Heppner are straightforward. Bradley Heppner, a criminal defendant facing federal charges in the Southern District of New York, input attorney case strategy materials into Claude, the AI platform built by Anthropic. Heppner did this on his own initiative, without direction from his defense counsel. During the execution of a search warrant, the government recovered those AI queries and the corresponding outputs.

The defense moved to suppress the AI interactions, arguing that the queries were protected under both attorney-client privilege and the work product doctrine. Judge Jed S. Rakoff rejected both arguments.

In an oral decision on February 10, 2026, followed by a written decision on February 17, 2026, Judge Rakoff held that no privilege or work product protection attached to Heppner's AI conversations. The court observed that "the implications of AI for the law are only beginning to be explored," but on the specific question before it, the answer was clear: a defendant voluntarily inputting materials into a consumer AI platform is not engaging in a privileged communication.

United States v. Heppner, No. 1:25-cr-00503(JSR), 2026 WL 436479 (S.D.N.Y. Feb. 17, 2026).

Why AI queries are not privileged communications

The legal reasoning behind the Heppner ruling rests on foundational principles of attorney-client privilege. Privilege attaches when a person seeks legal representation from an attorney, communicates in confidence for the purpose of obtaining legal advice, and there is a reasonable belief that an attorney-client relationship is forming or exists. Every element of that test fails when the counterparty is an algorithm.

AI platforms are not attorneys. They are software systems that process data and generate text based on pattern recognition across training corpora. No attorney-client relationship can form with a machine, regardless of how sophisticated its outputs may appear. Privilege does not attach at the moment someone searches for information -- it attaches when a person communicates with a licensed professional in a protected context.

Both the inputs and outputs of AI conversations are discoverable, in the same way that a Google search history is discoverable. There is no confidentiality expectation when a consumer submits queries to a third-party platform. Claude's own privacy policy gives Anthropic the right to disclose user data in connection with litigation, subpoenas, and legal proceedings. The voluntary disclosure of information to a third party with no obligation of confidentiality has long been held to waive privilege.

The court did note one narrow exception worth watching. Judge Rakoff observed that if Heppner's counsel had directed Heppner to use Claude as part of a legal strategy, the AI tool "might arguably be said to have functioned in a manner akin to a highly trained professional." But that is not what happened. Heppner acted on his own, and the unilateral use of a consumer AI tool carries no privilege.

Are AI chats admissible in court?

Yes. AI chat transcripts — both the queries a user submits and the responses the platform generates — are generally admissible and discoverable in litigation. The Heppner court treated AI interactions the same way courts have long treated search engine histories: as voluntarily created records with no expectation of confidentiality. Both federal and state courts have broad authority to compel production of electronically stored information, and AI conversations fall squarely within that category.

Can you get in trouble for conversations with AI?

The conversations themselves are not illegal. However, the content of AI interactions can be used as evidence in legal proceedings. In Heppner, the defendant's AI queries about his own criminal case strategy were recovered from his devices and deemed admissible. The risk is not in using AI — it is in assuming that what is entered into a consumer AI platform is confidential. It is not.

The Heppner decision is not an outlier. In January 2026, Judge Sidney Stein issued a ruling in In re OpenAI compelling broad discovery of AI training outputs, further establishing that AI-generated content occupies no special protected status in federal courts. The direction of the case law is consistent: AI conversations are discoverable, and courts are not inclined to extend privilege protections to interactions with software.

For advice on a specific situation, please talk to a licensed attorney in your state.

The privilege problem most legal tech ignores

The challenge with most legal technology platforms is that they blur the line between information and advice. They either deliver what functionally amounts to legal advice too early in the consumer interaction -- before any attorney-client relationship exists -- or they never create a clear moment of transition between the information phase and the representation phase. This ambiguity is where unauthorized practice of law risk lives. Not in the technology itself, but in the architecture around it.

Consider the scale of the problem. According to the Bureau of Justice Statistics, approximately 400,000 personal injury lawsuits are filed in the United States each year. Most settle before trial. The vast majority of those claimants navigate the early stages -- understanding whether they have a viable claim, what it might be worth, and which deadlines apply -- without any legal intelligence at all. They rely on advertising, word of mouth, or consumer AI tools that now, after Heppner, are confirmed to offer no confidentiality protection whatsoever.

The gap is not a technology problem. It is an architecture problem. The question is not whether AI can process legal information -- it obviously can. The question is whether the platform delivering that information is designed to keep consumers on the right side of the privilege line until the moment they choose to cross it by engaging an attorney.

One approach to the privilege gap: separating intelligence from advice

Caseworth operates entirely in the pre-representation phase. The platform delivers legal intelligence -- not legal advice. This distinction is not semantic. It is structural, and it is the foundation of Caseworth's compliance with Unauthorized Practice of Law rules across all jurisdictions where the platform operates.

The platform provides consumers with information about comparable outcomes, claim types, statutes of limitations, and benchmark ranges drawn from publicly available data. This is the UPL firewall: information (market benchmarks derived from public case data) versus advice (predictions about a specific person's case outcome or recommendations about what course of action to take).

Caseworth gives consumers the legal context they were never supposed to have access to. Not advice about their case. Information about cases like theirs.

When a consumer signals that they want representation -- that they are ready to move from intelligence to advocacy -- that expressed intent, combined with routing to a vetted attorney, is the beginning of a privileged relationship. It is not the end of a consumer product interaction. It is the start of something that carries legal protection. Everything before that moment is information delivery.

The intelligence lives in the pre-representation phase. The privilege attaches in the representation phase. The architecture keeps them separate.

The architecture is deliberate. Nothing in the Caseworth flow prior to the representation handoff could be characterized as legal advice. The Lexstimate report delivers benchmark ranges and comparable case data. The statute of limitations checker surfaces publicly available deadline information. The claim audit identifies claim types based on facts provided by the consumer. None of these features tell a consumer what to do. They tell a consumer what the landscape looks like.

Caseworth was designed from the ground up around UPL compliance principles. The engine delivers information — market benchmarks and comparable case data — not advice or predictions about any individual's outcome.

For advice on a specific situation, please talk to a licensed attorney in your state.

What the Heppner ruling means going forward

The Heppner decision is a first-mover ruling, not a final word. More courts will weigh in on AI conversation discoverability as the technology becomes more deeply embedded in legal workflows. Several implications are already taking shape.

First, closed enterprise AI systems -- where a law firm deploys an AI tool internally, keeps all data on-premises, and maintains strict confidentiality controls -- may be treated differently than consumer AI platforms. The Heppner court's emphasis on voluntary disclosure to a third party with its own data-sharing policies suggests that architecture matters. An AI tool deployed within a law firm's confidential infrastructure, used at the direction of counsel, occupies a different analytical framework than a consumer chatbot.

Should law firms add AI disclaimers to engagement letters?

Second, attorneys should begin including AI-specific disclaimers in engagement letters. Clients need to be warned — explicitly and in writing — that inputting privileged materials into consumer AI tools may waive privilege protections for those materials. A sample clause might state that the client agrees not to upload, copy, or input any attorney-client communications, case strategy documents, or privileged materials into any third-party AI platform without prior written consent from counsel. This is not speculative guidance. It is a direct consequence of the Heppner holding.

Third, clients and prospective clients should understand that anything entered into a consumer AI platform is potentially discoverable. This includes queries about legal strategy, case facts, and attorney communications that are copy-pasted into AI chat interfaces. The Heppner court drew no distinction between the substance of the query and the act of querying itself -- both are unprotected.

Fourth, the legal technology industry needs to take architecture seriously. The question is no longer just "what can our AI do?" It is "what is the legal status of the interaction between the consumer and our platform?" Platforms that cannot clearly answer that question are exposed -- not just to UPL risk, but to the possibility that their users' data will be used against them in litigation.

The access-to-justice imperative

The information gap in the justice system is real. Most people cannot afford to understand their own legal position. They do not know whether they have a viable claim. They do not know what comparable cases have settled for. They do not know which deadlines are running against them. And they certainly do not know that the AI tool they are using to answer those questions offers no confidentiality protection.

The solution is not giving legal advice cheaply. It is making consumers informed enough to seek representation with confidence. An informed consumer who understands their claim type, the relevant statute of limitations, and the general range of comparable outcomes is better positioned to find the right attorney, ask the right questions, and make informed decisions about their own case. That is the function of legal intelligence.

That is what a responsible approach to access to justice looks like. Intelligence first, counsel when it counts, and a clean boundary between the two.

The Heppner ruling makes the stakes clear. Every interaction a consumer has with an AI platform before engaging counsel is discoverable. Every query, every response, every uploaded document. The only responsible approach is to build platforms that treat the pre-representation phase for what it is: an intelligence function, not an advisory one. Caseworth was built on that principle. The Southern District of New York just confirmed why it matters.

For advice on a specific situation, please talk to a licensed attorney in your state.

Frequently asked questions

Are AI conversations protected by attorney-client privilege?

Generally no. In United States v. Heppner (S.D.N.Y. 2026), the court ruled that a defendant's queries to the AI platform Claude were not protected by attorney-client privilege or work product doctrine. AI platforms are algorithms, not attorneys, and privilege does not attach to interactions with them unless directed by counsel.

What is the Heppner ruling?

United States v. Heppner is a 2026 federal court ruling from the Southern District of New York holding that a criminal defendant's AI queries and responses were not protected by attorney-client privilege. The court reasoned that using a consumer AI tool is more like a search engine query than a confidential communication with an attorney.

Can the government access AI chat history?

Yes. AI conversations on consumer platforms like ChatGPT or Claude are generally discoverable in litigation. These platforms' privacy policies typically allow disclosure of user data in connection with legal proceedings. The Heppner ruling confirmed that such conversations do not carry privilege protections.

What is the difference between legal information and legal advice?

Legal information describes general legal principles, statutes, deadlines, and comparable case outcomes. Legal advice applies those principles to a specific person's situation and recommends a course of action. Platforms that provide legal information -- like Caseworth -- operate within UPL compliance. Platforms that cross into advice risk unauthorized practice of law.

How does Caseworth protect consumer data?

Caseworth's architecture separates the intelligence phase (legal information, benchmarks, deadlines) from the representation phase (attorney engagement). Consumer data collected during the intelligence phase is never characterized as privileged communication because no attorney-client relationship exists at that stage. When a consumer chooses to seek representation, the handoff is clean and deliberate.

Are AI chats admissible in court?

Yes. AI chat transcripts are generally admissible and discoverable in litigation. The Heppner court treated AI interactions the same way courts have long treated search engine histories: as voluntarily created records with no expectation of confidentiality. Both federal and state courts have broad authority to compel production of electronically stored information, and AI conversations fall squarely within that category.

Can you get in trouble for conversations with AI?

The conversations themselves are not illegal. However, the content of AI interactions can be used as evidence in legal proceedings. In Heppner, the defendant's AI queries about his criminal case strategy were recovered and deemed admissible. The risk is not in using AI — it is in assuming that what is entered into a consumer AI platform is confidential.

Should law firms add AI disclaimers to engagement letters?

Following the Heppner ruling, attorneys should include AI-specific disclaimers in engagement letters warning clients that inputting privileged materials into consumer AI tools may waive privilege protections. A sample clause should state that the client agrees not to upload attorney-client communications or case strategy documents into any third-party AI platform without prior written consent from counsel.

Educational analysis only · Not legal advice. This article analyzes a publicly available federal court ruling and describes Caseworth's platform architecture. It is for general informational and educational purposes only. This content does not constitute legal advice, does not create an attorney-client relationship, and should not be relied upon as a substitute for consultation with a licensed attorney. The legal landscape around AI and privilege is evolving rapidly. For advice on a specific situation, please talk to a licensed attorney in your state.