AI and Legal Privilege: Key Risks for Businesses

Tuesday 2 September, 2025

Last updated: 16 March 2026

We are aware that many of our clients are using generative AI services such as ChatGPT to assist with preparation of legal documents and communications. We recognise the benefits and efficiencies these tools can offer generally, and encourage our clients to use them. As a firm, Tompkins Wake is exploring AI initiatives that will drive operational efficiency and bolster our cybersecurity framework, within the framework of a responsible use policy.

However, in addition to the well-publicised inaccuracies that can be produced by generative AI, clients should be aware that their use of these services may not be caught by legal privilege under the Evidence Act 2006, as it presently stands in New Zealand. Under the Evidence Act, protection is provided where a client and their legal adviser communicate on a confidential basis regarding the provision of legal services, or where documents are created, compiled etc for actual or apprehended legal proceedings.  

While there are no New Zealand authorities at present on this question, in the United States, a judge has ruled that use of an open version of an AI tool by a non-lawyer meant the documents were not privileged. An England & Wales Tribunal has also stated that uploading confidential documents into open AI tools placed that information in the public domain, and therefore waived confidentiality and legal privilege.  

On that basis, if a client seeks legal advice by inputting information into the generative AI service (whether an open or closed environment), that information may be later found to be relevant as part of a discovery order, not protected, and be required to be disclosed in legal proceedings. This could have a negative impact on the client's position in those proceedings.  Likewise, inputting information for the purpose of providing the output to Tompkins Wake (or other legal advisors) may also not be privileged.

We also note that where clients' usage of generative AI is not in a closed environment (for example, a paid subscription), the information is likely to be used to train the AI system, which may impact confidentiality or copyright. The risks of utilising these services for legal advice were briefly visible when Google began indexing shared ChatGPT requests, making them searchable, before OpenAI moved to block this.  

For these reasons, we advise clients to be extremely cautious in their use of those services for legal advice. 

Our legal teams can help clients with this fast-moving area, including by:

• Preparing AI governance and usage policies tailored to the client’s risk profile;
• Reviewing contractual terms offered by AI vendors for compliance with NZ privacy and privilege requirements; and
• Advising on discovery or disclosure strategy when AI content is (or may be) in scope.
  
  

For further information, please don’t hesitate to get in touch with one of our experts below to discuss.