Introduction
Earlier this month, Sam Altman, Co-founder and CEO of OpenAI – the company behind the widely-used ChatGPT platform – made a public call for what he termed "AI Privilege". In this article we outline the events leading to that call. We then turn to the question of whether, under current New Zealand law, chat logs with generative AI tools are subject to legal advice privilege. We conclude that, in general, they are not. This means that those chat logs may need to be disclosed to another party in litigation or to a regulator in response to a request.
As generative AI tools continue to become more widely used for all manner of commercial purposes, what is outlined below emphasises the need for businesses to set clear policies and expectations as to the use of generative AI. In particular, when coupled with potential quality concerns with "legal advice" provided by a generative AI tool, in our view businesses should ensure that all legal advice continues to be sought and provided by an in-house or external lawyer.
New York Times v OpenAI
In December 2023, The New York Times filed a copyright infringement lawsuit against Microsoft, OpenAI and its affiliate companies, alleging the unlawful use of millions of its articles to train the large language model behind ChatGPT.
The claim sets out detailed allegations against the defendants, contending that the unauthorised use of the Times's articles "constitutes free-riding on The Times's significant efforts and investment of human capital to gather this information"1 and has caused economic harm through lost advertising and subscription revenue.2
The proceedings are intended to hold OpenAI and the other defendants responsible for billions of dollars in damages for the alleged unauthorised copying and use of the Times's "uniquely valuable works".3 The plaintiffs also seek an order requiring the destruction of any GPT, or other large language models, that incorporate the Times's works.4
OpenAI is defending the proceedings, including on the basis of the "fair use" doctrine, and has described the lawsuit as "baseless".5
As part of the proceedings, on 13 May 2025, the Court issued an order requiring OpenAI to preserve and segregate all output log data that would otherwise be deleted, effective immediately and until further order of the Court.6
Before the order, OpenAI's policy provided that, where users opted for a "temporary chat" or deleted their chat history, most data would be retained for 30 days before being permanently erased. Following the court order, however, OpenAI is now required to retain certain users' conversations with ChatGPT indefinitely.
In response, Mr Altman took to social media to advocate for the development of a broader legal and ethical framework protecting privacy in communications with AI. He adopted the term "AI Privilege" and argued that interactions with AI should be treated with the same confidentiality as those with a lawyer or doctor.
The intersection of generative AI and disclosure and privilege rules under New Zealand law
This raises the question: under New Zealand law, in what circumstances could parties be required to disclose legal questions posed by users to generative AI tools and the responses received?
The answer to that question starts with two basic concepts under New Zealand law:
- Under the High Court Rules, for example, the High Court can order discovery of "documents" that are or have been in a party's "control". Both concepts are very broadly defined: "documents" includes "information electronically recorded or stored, and information derived from that information"; and "control" means possession of the document, a right to possess the document, or a right to inspect or copy the document. There are numerous examples of the Court making orders requiring a party to litigation to disclose, for example, text messages or instant messaging chats stored on employees' devices.7 Regulators' compulsory production powers are similarly broad.
- Under the Evidence Act, legal advice privilege applies where a communication is:
- made between a person and their legal adviser (which includes someone enrolled and practising as a barrister and/or solicitor in New Zealand, and an overseas practitioner);
- intended to be confidential; and
- made for the purpose of obtaining or providing professional legal services.
It should not be assumed that user chats with generative AI tools are necessarily responsive to a discovery order or a regulator request. But if they are:
- they will fall within the meaning of "documents" under the High Court Rules; and
- legal questions posed and responses received will not usually be subject to legal advice privilege. At the risk of stating the obvious, even if the other elements of legal advice privilege were present, ChatGPT does not fall within the statutory definition of a "legal adviser".
There are some limited circumstances in which generative AI chats might be subject to privilege. For example:
- At common law, legal advice privilege extends to preparatory materials created in anticipation of seeking legal advice, even if not actually shared with a lawyer. The 2006 codification of the law under the Evidence Act arguably narrowed the common law position: on its face under s 54, the privilege applies only to communications that are in fact made to a legal adviser (the Law Commission has recommended a reinstatement of the common law position). In any event, the common law position remains relevant outside of the Evidence Act's application to a "proceeding". In these circumstances, where the generative AI chat was for the purpose of preparing a question for an (actual) legal adviser, it may be arguable that the chat is privileged.
- Preparatory materials privilege (also known as "litigation privilege") is broader than legal advice privilege and does not require the involvement of a lawyer. If proceedings are on foot or apprehended, a party can assert privilege under s 56 of the Evidence Act in relation to, for example, information compiled or prepared by the party for the dominant purpose of preparing for the proceeding. It is much easier to see how generative AI chats might meet this definition.
It is apparent that these are narrow circumstances. Unless and until Mr Altman's call for recognition of an "AI Privilege" gains traction (which seems unlikely, at least in a New Zealand context) legal advice should continue to be sought from an actual human lawyer.