US judge finds no privilege in AI-generated documents

The Heppner decision

In a first of its kind decision, a US judge ruled that documents generated by a client using an AI System (defined below) and later provided to a lawyer, could not be protected by legal privilege. The judgment, handed down on 10 February 2026 in the Southern District of New York, considered whether materials created by the defendant, Mr Heppner, using a public AI System to prepare for a fraud investigation, could be withheld from disclosure. The judge was clear: applying established US law, documents generated by a generative AI System at the request of a non-lawyer were not protected by privilege on a number of grounds, and such documents could not become privileged merely because they were subsequently shared with a lawyer in relation to anticipated litigation.

Our US colleagues have considered the decision in their article: Are AI-generated documents protected from discovery if you send them to your lawyer? One judge says “no”. Here we consider the case from an English law perspective, including how the principles may be determined by English courts.

We expect that the relationship between legal work generated by or with the assistance of AI and legal professional privilege (LPP) will come before the English courts in the near future, and absent any current substantive authority on the issue, the judge’s findings in Heppner have the potential to be persuasive. The judge's analysis of confidentiality, US law "attorney-client privilege" and the "work product doctrine" offer valuable guidance and can inform best practice when using generative AI in a legal context. We draw out key parallels – and areas of divergence – below. Our analysis focuses on the following key issues:

• Can an AI System give privileged legal advice?
• Are communications with an AI System confidential?
• What are the analogous implications of the judge's findings in Heppner for English law litigation privilege?

For full analysis of the interaction between generative AI and the English law position on privilege, read our full report: Is Legal Advice Developed by or with an AI System Legally Privileged?

 

Key takeaways

• The judge in Heppner endorsed a "first principles" approach to generative AI and privilege: "AI's novelty does not mean that its use is not subject to longstanding legal principles". Similarly, when considering privilege and AI from an English law perspective, the prudent approach is to apply existing common law principles.

• An AI System cannot itself give privileged legal advice – whether in the context of litigation or otherwise; there is no "AI Privilege".
• Users of AI Systems should ensure inputs and outputs remain confidential. If system providers warn otherwise, privilege arguments are likely to be impossible. Users should check the terms and conditions of AI Systems closely and only input confidential information into AI tools with adequate confidentiality safeguards (likely through bespoke terms in a licencing agreement or use of closed private AI Systems).
• In contrast to US law, confidential use of an AI System by a lay client to produce material for subsequent discussion with their lawyer about pending litigation may be capable of protection under the English law principles of litigation privilege, provided all other criteria are met.

 

AI Privilege

In Heppner, the judge made clear that AI Privilege does not exist as a standalone concept: "because [the AI System] is not an attorney… that alone disposes of [the claim] of privilege". The position is analogous under English law: "legal advice" generated by an AI System and provided directly to a non-lawyer is not capable of being privileged, as the AI System is not a lawyer. It would be for Parliament to extend the scope of legal advice privilege (LAP) beyond legal advice given by legal professionals.

The judge in Heppner did, however, consider an alternative position under US law: if counsel for Heppner had directed Heppner to use the AI System, "[it] might arguably be said to have functioned in a manner akin to a highly trained professional who may act as a lawyer's agent within the protection of the attorney-client privilege".

As we discuss in section 7 of our full report, "AI Systems and legal advice privilege: Could an AI System ever be regarded as a lawyer?", the English courts have adopted a narrow interpretation of "agent" for LPP purposes. Under LAP, a client or their legal adviser can communicate through an agent, but for such communication to attract privilege, that agent "must be no more than a means of communication"¹ or "alter ego" for either the client or the lawyer. There is no need for an agent to be a "highly trained professional" – merely a conduit for the communication between the lawyer and their client. The autonomous characteristics of generative AI and their enhanced "intelligence" (even if artificial) - which distinguishes them from basic search engines or word processing programmes – means that, in our view, a court is likely to see AI as more than a mere agent, with the result that communications with AI could not generate LAP.

The prudent approach is for AI System to be treated as analogous to subordinates of a lawyer for LAP purposes. In practice, this means that the lawyer should thoroughly review, verify, and where appropriate, amend legal advice generated by an AI System, and issue final work product in the name of the lawyer who has checked the output. See our article Should an AI system be treated like a trainee lawyer or paralegal? | DLA Piper for further detail.

 

Confidentiality

For a communication to be protected by LPP under English law, it must be (and remain) confidential². As we discuss in section 5 of our full report, "Confidentiality and privilege: How does confidentiality apply to AI Systems?", a court in this jurisdiction is likely to start from the same position as that adopted in Heppner, that inputs into free versions of popular AI Systems (Public AI Systems), are unlikely to be confidential. This is because Public AI Systems may use data to train an underlying model, and users are typically unable to negotiate the terms of service or licence.

Certain Public AI Systems may also actively disclaim to users that their data – specifically, inputs and outputs generated during their interaction with the AI System – may be disclosed to third parties. This was the case for the AI System in Heppner: the relevant privacy policy explicitly stated that the AI company reserved the right to disclose users' interactions with its chat tool to third parties, including governmental and regulatory authorities. According to the judge in Heppner, that "clearly puts [the AI System's] users on notice that… [the AI company], even in the absence of a subpoena compelling it to do so, may "disclose personal data to third parties…". The judge also cited the New York case of re Open AI, Inc, Copyright Infringement Litig.³, in which it was held that AI users "do not have substantial privacy interests"⁴ in their conversations with Public AI Systems. In that litigation, OpenAI was ordered to disclose 20 million redacted user interactions with ChatGPT.

English law could potentially allow exceptions to the general position that communications with a Public AI System are not confidential, for example where a party can demonstrate that the loss of confidentiality is only theoretical. However, given the hard line adopted in Heppner and OpenAI, and widely publicised warnings about the confidentiality of Public AI Systems (including in the judicial guidance note on the use of AI)⁵ such arguments are unlikely to provide a reliable fallback. Indeed, the English courts have recently held that placing information into an open source AI tool was to place that information on the internet in the public domain and thus to breach client confidentiality and waive legal privilege.⁶ The prudent approach is to assume that information entered into such systems will be regarded as "published to all the world". Instead, users should opt for closed or private AI Systems for any confidential communications.

 

Litigation privilege

In a parallel to English law, the judge in Heppner restated the position that "it is black-letter law that non-privileged communications are not somehow alchemically changed into privileged ones upon being shared with counsel". Accordingly, as Heppner's communications with the AI System were not capable of being privileged in the first place, they could not become privileged simply by being shared with a lawyer. The same principles apply under English law. But what if they are prepared and provided to a lawyer when litigation is in contemplation?

In Heppner, the judge considered whether in view of the fraud investigation, the AI documents could be protected by the "work product doctrine" which has similar considerations to English law litigation privilege principles. The judge determined that even though the AI documents were created in anticipation of litigation, they could not be privileged as they were prepared "by the defendant on his own volition". Notably, the analysis under English law would be different.

Litigation privilege can extend to certain types of confidential documents which are not strictly communications, but are created by a lawyer, a third party, or (as in this case) a client, and which come into existence to enable legal advice to be sought or given, or information or evidence to be prepared or gathered, in relation to the litigation⁷ which has to be in reasonable contemplation.

Heppner used the AI System to assist with instructions to his lawyer after he became aware that he was being investigated for fraud. Under English law, litigation would, in all likelihood, have been considered in contemplation at the time of Heppner's interaction with the AI System. Provided Heppner could prove that the dominant purpose of the interaction with the AI System was to inform preparations for his defence, then such inputs and outputs - even while not strictly communications at the time they came into existence – would likely satisfy the test for protection under litigation privilege. In that respect, the decision in Heppner leaves open the possibility for an English judge to reach a different conclusion with respect to litigation privilege in a similar scenario, provided, importantly, that a closed AI System was used and the use of AI did not compromise confidentiality.

 

Closing thoughts

The English courts will no doubt need to grapple with issues similar to those presented in Heppner at some point in the future, and when they do, they will have the benefit of considering the reasoning taken by the US judge. It will be interesting to see if a similar approach of applying long standing legal principles is followed.

---

¹ Sir George Jessel MR in Wheeler v Le Marchant (1881) 17 Ch D 675 at (651).
² Glencore International AG v Commissioner of Taxation (2019) HCA 26.
³ No. 25 MD 3143, EFC No. 1021 at 3 (Jan 5, 2026).
⁴ Confidentiality and privacy are not treated as interchangeable under English law. See Is legal advice developed by or with an AI System legally privileged? | DLA Piper for further detail.
⁵ See Courts and Tribunals Judiciary, "Artificial Intelligence (AI) Guidance for Judicial Office Holders", dated 31 October 2025.
⁶ UK and R (on the application of Munir) v Secretary of State for the Home Department [2026] UKUT 00081 (IAC)
⁷ See Passmore, Privilege (Fifth Edition), Thomson Reuters, 2024, para. 3-013.