AI and Privilege: Lessons from U.S. v. Heppner and the Emerging UK Position

In February 2026, United States v. Heppner, 25 Cr. 503 addressed whether AI-generated defence analysis is protected by the attorney-client privilege or if it is work product. Meanwhile, the English courts have engaged the issue from a different angle, confronting professional duties and accuracy risks following the use of AI as seen in Ayinde v. London Borough of Haringey and Al-Haroun v. Qatar National Bank [2025] EWHC 1383 (Admin). More recently, in UK v. Secretary of State for the Home Department [2026] UKUT 81 (IAC), the Upper Tribunal went further, expressing the view that uploading confidential client material to a public AI platform could amount to placing it in the public domain, thereby breaching confidentiality and waiving privilege.

The United States: Privilege, Waiver and Supervision
On February 10, 2026, the Judge in the Heppner case ruled that the AI-generated documents in this matter were not protected by attorney-client privilege or by the work product doctrine.

The defendant in this matter, Heppner, was aware that he was the subject of an investigation, and used the public generative AI tool, Claude, to generate written exchanges analyzing potential defence strategies and arguments. On November 4, 2025, FBI agents executed a search warrant at Heppner’s residence and seized various documents and electronic devices. Among the devices recovered were 31 documents recording communications between Heppner and Claude.

The government sought a ruling that the AI-generated documents were not protected by attorney-client privilege or by the work product doctrine. Heppner’s defence counsel claimed privilege over the materials on the basis that Heppner created the documents to discuss and share with his defence counsel. Importantly, his counsel conceded that they had not instructed Heppner to undertake this task, nor had they supervised Heppner, and that ultimately, Heppner had conducted the use of the AI tool of his own accord.

Judge Rakoff granted the motion from the bench and subsequently issued a written memorandum explaining that:

1. The exchanges were communications between Heppner and Claude and were not communications with an attorney.
2. That privilege requires confidentiality and that Heppner did not have a reasonable expectation that his communications with the AI platform would remain confidential, given the platform’s disclosures and terms.
3. Privilege attaches where communications are made for the purpose of obtaining legal advice from a lawyer.

The court also rejected the argument that later transmission of the AI outputs to counsel could retrospectively confer privilege. A non-privileged communication does not become privileged because it is subsequently shared with a lawyer. The court also addressed waiver. Even if some of the information input into the AI system had been privileged when originally communicated between lawyer and client, disclosure to the AI provider constituted waiver. This aspect of the decision underscores the practical risk that entering privileged material into a public generative AI system may undermine its protected status. Heppner also relied on the work product doctrine. The court assumed that the documents were prepared in anticipation of litigation but held that they were not prepared by counsel nor prepared at counsel’s direction. Because Heppner acted independently and not as counsel’s agent, the work product doctrine did not apply.

Importantly, the decision was fact specific. The ruling is not stating that AI outputs are never privileged, but it does demonstrate that unsupervised client use of a public AI tool is unlikely to satisfy privilege or work product requirements.

England and Wales: Confidentiality, Accuracy and Professional Responsibility
While there is not yet direct English authority equivalent to Heppner addressing privilege over AI outputs in those terms, English courts and regulators have engaged extensively with generative AI in litigation, particularly where incorrect AI outputs have been submitted to the court.

AI Hallucinations and Professional Duties
In Ayinde and Al-Haroun, the court dealt with cases in which documents had been relied upon by the Claimants which contained inaccurate or fabricated authorities arising from the use of AI tools. The court emphasized that legal representatives have a duty to ensure that submissions and authorities are accurate. The fact that material may have been generated with the assistance of technological tools does not diminish that responsibility. Failure to verify AI-generated material may expose professionals to consequences, including wasted costs orders and referrals to relevant professional bodies.

More recently, the Upper Tribunal in UK v. Secretary of State for the Home Department heard joint cases in relation to matters whereby fictitious authorities had been submitted following the use of AI-generated documents. For the first matter, the Tribunal expressed the view that uploading confidential client emails and letters to an open AI tool amounted to placing the material in the public domain, thereby waiving legal privilege and breaching confidentiality. Although no referral was made due to self-reporting, the Tribunal emphasized that failure to verify AI-generated citations would ordinarily justify regulatory referral. In relation to the second matter, it was noted by the Tribunal that a qualified legal practitioner who delegates work to a colleague retains ultimate responsibility for supervising work and ensuring its accuracy. The Tribunal referred the solicitor to the Solicitors Regulation Authority.

Professional guidance in England and Wales reflects the same themes. Judicial and regulatory materials stress the importance of preserving confidentiality, maintaining effective supervision, and ensuring that work product is independently reviewed before being relied upon.

In England and Wales, the judiciary’s focus has been on accuracy, verification and duty to the court. However, the underlying principle is similar to Heppner: AI is a tool, and responsibility remains with the human professional.

Conclusion
In the U.S., Heppner establishes that generative AI is not automatically a confidential intermediary for privileged communication. In England and Wales, the Upper Tribunal in UK v. Secretary of State for the Home Department expressed a firm view on uploading documents to an open AI tool and the potential impacts on confidentiality and waiver. English law recognises legal advice privilege and litigation privilege, both of which depend fundamentally on confidentiality. Where confidentiality is lost, privilege is normally lost with it.

In practical terms, the position is unlikely to be absolute. Where a system is deployed under clear legal direction, with appropriate safeguards and oversight, the privilege analysis may be more fact sensitive. At the same time, recent decisions demonstrate that courts will not tolerate unverified and incorrect AI-generated output. The consistent emphasis on confidentiality and professional responsibility with regard to the use of AI tools suggests that outcomes similar to Heppner could arise in England and Wales if privileged material is freely shared with a third-party AI system without adequate protection.

One practical takeaway is that, however tempting, clients should be very wary of seeking a “second opinion” from any public AI tool, including uploading confidential documents, and worse still, any legal advice received, as there is a real danger of privilege being waived.