The rapid integration of artificial intelligence into legal practice has created a new and underappreciated risk: the inadvertent waiver of legal professional privilege. Recent judicial developments are beginning to illuminate the fault lines. Practitioners need to keep on top of these developments and advise their clients accordingly.
Privilege
Whilst there are other types of privilege, in this article we focus on legal professional privilege (“LPP”) comprised of:
- Legal advice privilege: covering confidential communications between lawyers and their clients made for the dominant purpose of seeking or giving legal advice; and
- Litigation privilege: covering confidential communications between lawyers and clients, or between the lawyer or client and a third party, which come into existence for the dominant purpose of being used in connection with actual or pending litigation.
Confidentiality is key to both definitions; a communication which is not confidential cannot be privileged.
Privilege & AI: Where are we?
There is not yet authority from the Employment Tribunal on the interaction between AI and legal professional privilege. However, the recent combined judgment in UK v Secretary of State for the Home Department [2026] UKUT 81, in the Upper Tribunal Immigration and Asylum Chamber, shed some light on judicial thinking in this area. Although the primary focus of the judgment was on legal representatives having relied on fictitious authorities generated by AI, the Judge went further, offering a view on the intersection of confidentiality, privilege and AI.
“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 thus to breach client confidentiality and waive legal privilege, and any such conduct might itself warrant referral to the SRA and should, in any event, be referred to the Information Commissioner’s Office.” The Judge also added that “Closed source 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.”
It must be acknowledged that the Judge's use of the terms "open source" and "closed source" depart from their established meanings in the technology sector. In the AI sphere, "open source" typically refers to freely available source code released openly to the public, whereas “closed source” models are proprietary systems that keep their code confidential. The distinction the Judge appears to be alluding to is between publicly available consumer tools and private enterprise tools — but the terminology matters, and practitioners should not assume that an "open source" model in the technical sense is synonymous with one that shares user data publicly.
Importantly, the judgment indicates that where privileged information is uploaded to a publicly available tool such as ChatGPT or Claude, the tool may use the input and/or output data to further train its model, improve it, or provide it to third parties, meaning confidentiality is not preserved and privilege may be lost.
Can AI-generated legal advice attract privilege?
A separate question is whether obtaining legal advice from an AI tool can attract LPP in the first place? This question has been addressed by the courts in the United States. In United States v Heppner, 25 Cr. 503 (JSR), the defendant, facing fraud charges, had used the AI platform Claude to prepare documents outlining his defence strategy, which he subsequently shared with his lawyers, claiming privilege over them. The court rejected the privilege claim on multiple grounds including those below.
- Claude is not an attorney; because the documents were not communications between an individual and their attorney, there was no privilege.
- The communications were also not confidential Not merely because the individual had communicated with a third-party AI platform, but because Claude's written privacy policy expressly provided that Anthropic collects data on both users' inputs and Claude's outputs, uses such data to train its model, and reserves the right to disclose such data to third parties.
- Furthermore, the individual had not communicated with Claude for the purpose of obtaining legal advice.
Whilst US privilege concepts do not map precisely onto English law, the decision serves as a reminder that clients who use AI platforms to assist their own legal thinking, particularly without their lawyers' direction, may inadvertently place sensitive material beyond the reach of any privilege protection, rendering it potentially disclosable.
The message for practitioners advising employers is clear. Any arrangement in which an HR team or in-house counsel uses a consumer-facing AI tool to develop legal strategy, without meaningful external lawyer involvement, risks generating documents that are neither privileged nor confidential.
Expansion of privilege?
The High Court decision in Aabar Holdings S.A.R.L & Others v Glencore PLC, Ivan Glasenberg and Steven Frank Kalmin [2026] EWHC 877 suggested a potential expansion of legal advice privilege to cover "intra-client" communications (i.e., internal communications between members of the client group).
The Judge held that a party is entitled to assert legal advice privilege in respect of intra-client documents, provided that those documents were created with the dominant purpose of seeking legal advice. If upheld and followed, this would expand the scope of privilege to capture communications and documents created by or exchanged between members of the client group for the dominant purpose of seeking legal advice (even where no lawyer is directly involved in the communication), rather than solely covering communications between lawyers and their clients.
As a High Court decision, this is not binding on other courts. Whilst the judgment does not address AI directly, looking at the concept through an AI lens, the prudent advice remains that clients using generative AI to create documents with the dominant purpose of seeking legal advice should exercise caution and continue to work on the basis that the inputs and outputs would not attract legal advice privilege.
Practical steps
Practitioners advising employer clients should consider the following framework of practical steps.
- Educate and train. Staff should understand the issues at play; if they input privileged material into publicly available AI tools, privilege is likely to be lost. Even when using enterprise AI tools, care must be taken to ensure privilege is maintained.
- Review policies. Regular review and refresh of AI usage policies, IT policies and confidentiality policies is advisable to cover points which emerge as AI use evolves, and to ensure robust guidelines are in place.
- Conduct supplier due diligence. Employers should carefully examine the contractual terms of any AI tool or service provider before deployment, paying particular attention to data ownership, confidentiality obligations, and whether data is fed back into the provider's model for training purposes.
- Keep reviewing. AI tools and supplier terms evolve rapidly. Employers should build in a process for regular review to ensure arrangements remain legally compliant and fit for purpose as the technology and regulatory landscape develops (including taking advice from data protection specialists to ensure compliance with data protection regimes and any EU AI Act obligations).
- Maintain lawyer involvement. Maintaining lawyer oversight of AI-generated legal advice is crucial, not only to avoid hallucinations and incorrect advice, but to help attract legal professional privilege when it should rightly apply.
Copyright & permissions
.jpg&w=256&q=75)
.png&w=3840&q=75)
.jpg&w=3840&q=60)
