knowledge 29 April 2026 |

Navigating Legal Privilege when using Generative AI

Recent rulings on both sides of the Atlantic have highlighted the growing tensions between the use of generative AI tools and the preservation of privilege.  The decisions also serve as a warning to legal practitioners that materials generated with such tools may not be covered by privilege.  We examine these developments and outline some practical measures which Irish in-house lawyers can adopt to reduce the risk of inadvertently losing privilege protection when using AI.

What is privilege?

Privilege refers to the protection that shields certain communications from being produced in litigation.  There are a number of different forms of privilege.  Legal Professional Privilege (“LPP”) protects the confidentiality of lawyer client communications and is divided into legal advice privilege (“LAP”) and litigation privilege.  Whereas LAP protects confidential communications between a lawyer and client for the purpose of giving or receiving legal advice, litigation privilege protects confidential communications between a client and lawyer, or between either of them and a third party, made for the dominant purpose of preparing for actual or reasonably apprehended litigation.

LPP can be waived, meaning that the protection of privilege can no longer be relied upon.  While we await Irish authority, recent court rulings in the UK and US highlight that the use of generative AI tools, in certain circumstances, may result in a waiver of privilege.

UK

On 17 November 2025, the UK’s Upper Tribunal of the Immigration and Asylum Chamber observed that uploading confidential documents into an open-source AI tool such as ChatGPT placed this information in the public domain, and so breached client confidentiality and waived legal privilege. The decision (available here) is possibly the first reported case in England and Wales to consider the impact on privilege of using public AI tools.  The Court also noted 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 the risks outlined above, namely breaching client confidentiality and waiving legal privilege.

US

We understand that to date in the US three different federal courts have addressed the question of how privilege and/or the work product doctrine applies when parties use generative AI tools in connection with litigation.

In United States of America v. Heppner¹ Bradley Heppner was charged with securities and wire fraud.  After receiving a grand jury subpoena but before his indictment he used the publicly available version of Anthropic’s Claude to prepare 31 documents related to his legal case, which he shared with his counsel (the “AI Documents”).  When the AI Documents were uncovered by the FBI on Mr Heppner’s computer the US District Court in the Southern District of New York was asked to rule on whether the materials were protected by either attorney-client privilege or the work product doctrine².   The Court found that they were not.

The AI Documents lacked the essential elements of attorney-client privilege (which is similar to LAP under Irish law).

• The AI Documents were not communications between Mr Heppner and his counsel.
• The communications memorialized in the AI Documents were not confidential.  This is not merely because Mr Heppner communicated with a third-party AI platform but also because the written privacy policy to which users of Claude consent provides that Anthropic collects data on both users’ “inputs” and Claude’s “outputs”, that it uses such data to “train” Claude, and that Anthropic reserves the right to disclose such data to a host of “third parties” including “governmental regulatory authorities”.  Therefore, Mr Heppner could have had no reasonable expectation of confidentiality in his communications with Claude.
• The court was not persuaded that Mr Heppner had communicated with Claude for the purpose of obtaining legal advice, commenting that the question was whether he intended to obtain legal advice from Claude.  The Court also noted that Claude disclaims providing legal advice. 

The Court also found that the documents did not fall within the work product doctrine (which is similar, albeit not identical to, litigation privilege under Irish law) as the AI Documents were not prepared at counsel’s direction and did not reflect the defence strategy at the time of their creation.

Under Irish law, for litigation privilege to be successfully asserted over a document, there is no requirement that the document in question be created by or at the direction of a lawyer.  Therefore, setting aside the issue of confidentiality an Irish Court might potentially have reached a different conclusion in this case that litigation privilege applied if satisfied that the AI Documents were made for the dominant purpose of preparing for actual or reasonably apprehended litigation.

In two further US decisions, the US district courts in Michigan and Colorado in Warner v Gilbarco Inc and Morgan v V2X, Inc (“Morgan”) have held that self-represented parties bringing civil proceedings were entitled to claim privilege, in the form of the US work product doctrine over their interactions with a public AI tool.  The findings that privilege was not waived despite the use of a public AI tool seemingly contrasts with the UK decision (considered above) and the approach of the Court in Heppner.  The decision in Morgan is also interesting in ordering the restriction of the use of AI in respect of confidential documents disclosed by the opposing party, essentially prohibiting inputting confidential information into a public AI tool, and in requiring the party who had used a public AI tool to disclose the name of that tool.

Despite these decisions, parties should exercise caution in their use of AI in the delivery of legal services.  In particular:

• Understand how the AI tool may use your information – if you upload confidential documents or input confidential information into some AI tools this may result in putting this information in the public domain, breaching client confidentiality and waiving legal privilege.  Review the relevant terms and conditions to understand how the AI tool may use your confidential information.
• Use enterprise systems over public AI tools – where a public AI tool is used, there is a significant risk that privilege will not be available for new material created by that AI tool, or that privilege in materials uploaded to the public AI tool will be lost. 
• Make sure that the elements of LPP are complied with – sight cannot be lost of the fact that the elements of LPP discussed above still need to be complied with for a claim of privilege to be successfully asserted.  For example, if litigation is not in prospect, then the communication must be for the purpose of giving or receiving legal advice.
• Ensure guidelines are in place on the use of generative AI – guidelines should be in place on the appropriate use of AI by employees, in particular warning against the use of public AI tools.

Recently, the Irish Court of Appeal in Guerin v O’Doherty [2026] IECA 48 (linked here) set out principles of general application in relation to the use of AI in litigation.  While the Irish Courts have yet to grapple with the question of privilege and AI, it seems inevitable that this question will raise its head in the coming months and years.