The increasing use of generative artificial intelligence (Gen AI) tools by businesses means the risks of losing confidentiality and legal professional privilege over documents need to be managed. In the recent case of UK v Secretary of State for the Home Department [2026] UKUT 81 (Hamid), the Upper Tribunal (Immigration and Asylum Chamber) delivered the first decision by an English court or tribunal to directly address the implications for legal professional privilege when confidential and privileged material is uploaded to open-source AI tools (such as ChatGPT).
The Upper Tribunal confirmed that uploading documents to open-source AI tools will breach confidentiality and waive legal professional privilege, drawing a distinction between open-source AI tools and closed source AI tools (which operate within a secure network). Although Hamid arose in the immigration context, the principles it discusses will have significant relevance for any organisation that handles privileged or confidential information: solicitors and clients should both be aware of the risks of using open-source AI tools.
For further information on this topic, we discussed the interplay between Gen AI and legal professional privilege and other risk areas in more detail in our earlier article: Legal professional privilege in the generative AI era.
Background
The Hamid case joined together two Tribunal cases relating to the use of AI by solicitors which resulted in false and inaccurate case citations being placed before the Tribunal:
- In the first matter, a regulated immigration adviser admitted to having used ChatGPT in the course of his legal work. He accepted that the fictitious citation he had referred to was an AI hallucination, although he maintained that this had occurred unknowingly. The adviser explained that he had used ChatGPT to improve emails to clients and had also uploaded Home Office decision letters into ChatGPT in order to summarise them for clients. The adviser had self-reported to the Immigration Advice Authority and the Solicitors Regulation Authority (SRA).
- In the second matter, the legal representative denied having used ChatGPT, contending instead that the document had been drafted by a trainee based on an outdated precedent and practitioner notes. The Tribunal examined how the inaccurate citations arose and whether referral to the relevant regulatory body was appropriate. The Tribunal ultimately referred the solicitor to the SRA.
The Upper Tribunal emphasised that legal professionals are obliged to ensure the accuracy of documents placed before courts and tribunals, and referred to the guidance provided in R (on the application Ayinde) v London Borough of Haringey [2025] EWHC 1383 (Admin). It noted that legal professionals who delegate their work remain responsible for the supervision and accuracy of the work product and are, "expected to ensure that such documents are checked [and] that errors are identified." Failure to conduct such checks could result in serious disciplinary consequences.
The implications for legal professional privilege
Beyond the question of fictitious citations, the Upper Tribunal used the Hamid decision to provide clear guidance on the implications of using AI tools on confidential and legally privileged documents and information.
The Tribunal drew a clear distinction between publicly available AI tools (referred to as "open source" tools, such as ChatGPT) and closed AI systems (such as Microsoft Copilot). In summary, any confidential or privileged information will be considered public if it is uploaded into an open AI platform, waiving privilege and potentially breaching any confidentiality obligations. The Tribunal held that “uploading confidential documents into an open source AI tool such as ChatGPT is to place this information on the internet in the public domain and […] waive legal privilege”.
The Upper Tribunal acknowledged that closed AI tools "which do not place information in the public domain, such as Microsoft Copilot, are available for tasks such as summarising without these risks." This distinction between open-source and closed-source AI tools is therefore crucial to the question of whether confidentiality and legal privilege over documents has been waived. However, although not discussed by the Tribunal, even when using closed AI tools, businesses should take care with privileged materials and be mindful of the existing principles of legal privilege. Materials containing legal advice, including prompts and AI output as well as more traditional types of documents, should be stored securely and access restricted to those who need it, to ensure they remain confidential and privileged. We consider the use of closed AI tools in our earlier article, as mentioned above.
The Tribunal’s approach is consistent with recent guidance (available here) provided by the Courts and Tribunal Judiciary of England and Wales, which stated that information uploaded into a public AI chat bot “should be seen as being published to the world”. This case now confirms this guidance and warns of the risks of open-source AI.
Similarly, in a recent US decision, United States v Heppner (S.D.N.Y., February 2026), a US federal court held that documents created using a public AI platform were not protected by attorney-client privilege, and the reasons for the decision included that neither the prompts nor the outputs were confidential (see our US team’s article: Recent GenAI rulings highlight challenges with safeguarding privilege). Although the cases considered slightly different issues (in Heppner, the court concluded that privilege never arose, whereas in Hamid, the focus was on waiver of privilege that had previously existed) the direction of travel is clear.
Key takeaways
Although the decision of the Upper Tribunal does not bind other courts and tribunals, it is likely to be at least persuasive. It also provides clear guidance for businesses and legal professionals:
- Confidential and privileged information should not be uploaded to open-source AI tools to avoid waiving confidentiality and legal professional privilege, which would otherwise protect documents from disclosure.
- If privilege over a document is waived by the use of an open-source AI tool, it is no different to waiver by other means: i.e. there is no mechanism by which privilege can be retrospectively asserted or recovered.
- Where AI tools are used for legal tasks, organisations should use enterprise-grade, closed AI systems with appropriate contractual and technical safeguards that prevent disclosure of information or the use of input data for training purposes. However, as noted above, even when using closed AI tools, businesses should ensure that prompts and outputs containing privileged information are stored securely and access to any materials containing legal advice is restricted to those in the business who need it (in the same way they should already be doing with other types of privileged documents).
- The use of AI does not form an exception to the principle that legal professionals, whether in-house or in private practice, remain fully responsible for supervising work they have delegated. Failure to check the accuracy of documents prepared with AI assistance that are put before a court may lead to regulatory referral and serious disciplinary consequences.
- Businesses should review their AI usage policies and governance frameworks to ensure the adequacy of staff training on privilege and confidentiality in the context of AI. This case reiterates the importance of sufficient training and an awareness of the nature of the AI tool being used in order to avoid the potential pitfalls that may arise with the use of AI.
With thanks to Emma Torok for her assistance in preparing this post.
Subscribe to our Inside Disputes Blog to receive regular expert insights into banking and commercial litigation in the English courts and issues relevant to investigations.