Simpson Grierson - AI and legal privilege - Does using AI put privilege at risk?

AI and legal privilege - Does using AI put privilege at risk?

A lot has been said recently about the possible benefits of AI when it comes to legal advice and litigation. What’s less clear is what happens when legally privileged information is entered into an AI tool, or a person goes to AI for legal advice.

Can legal privilege be claimed? Has it inadvertently been lost by using an AI platform? While these questions remain untested in New Zealand, overseas courts have started to grapple with them. Those decisions provide some useful indications as to how the New Zealand courts may approach these issues, and serve as a warning that the benefits of AI can quickly be undermined if privilege considerations are not carefully managed.

Key takeaways

The impact of AI on legal privilege is an unsettled area: We are yet to see a major decision in New Zealand and the courts overseas have reached different conclusions about the underlying issues. Further, the different outcomes are often fact-specific.
Confidentiality is likely to be a critical issue: In New Zealand, legal privilege depends heavily on confidential communications remaining confidential. Uploading sensitive information into publicly available AI tools may jeopardise privilege - particularly where the relevant terms and conditions permit the collection, use, or disclosure of inputs and outputs.
AI won’t qualify as a “legal advisor” for the purposes of the Evidence Act 2006: While AI tools may do a good job of sounding like a lawyer, communications with them won’t attract legal advice privilege.
Organisations should proceed cautiously: Until the law becomes clearer, organisations and individuals should think very carefully before inputting privileged, confidential, or dispute-related material into AI tools without appropriate legal advice and support. Policies governing AI use should reflect this. Put simply - the efficiency gains offered by AI can be quickly outweighed if privilege is lost and sensitive information becomes disclosable in subsequent proceedings.

Overseas cases addressing AI and legal privilege

There are now a handful of cases from the United Kingdom and the United States that determine how legal privilege applies to AI inputs and outputs. This list of cases is only likely to grow as the use of AI in a legal setting increases.

United States v Heppner - United States District Court, Southern District of New York

The Defendant in this case had been charged with various fraud and dishonesty offences. A search warrant for his property resulted in the seizure of electronic devices, and the discovery of exchanges between the Defendant and the AI programme Claude regarding his legal position and litigation strategy. These were subsequently shared with his lawyers.

The Court held that privilege did not attach to the relevant inputs and outputs on the basis that:

Claude is not a lawyer and was not able to give actual legal advice. Claude itself disclaimed any ability to give legal advice.
The communications were not treated with sufficient confidentiality, as the relevant version of Clause was public and its terms and conditions allowed for the collection, use, and potential disclosure of inputs and outputs.

Warner v Gilbarco Inc - United States District Court, Eastern District of Michigan Southern Division

This was an employment discrimination case. The Plaintiff represented herself and used a publicly available version of ChatGPT to assist in drafting legal documents and provide related legal analysis. The Defendant sought disclosure of the relevant material by claiming it was not legally privileged because it had been shared with ChatGPT and was accordingly no longer confidential.

The Court declined the Defendant’s request by holding that the material was subject to the equivalent of litigation privilege (which covers material that is prepared for the dominant purpose of proceedings that are in reasonable contemplation). In particular, the Court held that the Plaintiff had not waived privilege because any third-party waiver needs to occur to a person and that AI programmes are “tools, not persons”. It also held that a waiver only arises where the relevant disclosure occurs to an adverse party.

Morgan v V2X Inc - United States District Court, District of Colorado

This was another employment discrimination case with a self-represented Plaintiff. The employer sought disclosure of details regarding what AI programmes the Plaintiff had used in connection with its confidential information.

The Court declined the request and allowed the Plaintiff to maintain privilege in its AI interactions. In doing so, it followed the reasoning of Warner, and distinguished Heppner on the basis that different rules applied in civil proceedings and the fact that the Plaintiff was effectively acting as his own lawyer.

UK v Secretary of State for the Home Department - United Kingdom Upper Tribunal - Immigration and Asylum Chamber

Following two instances of lawyers submitting AI-generated materials containing false citations, the Tribunal used these cases as an opportunity to set out the obligations for legal professionals using AI. Amongst other things, it confirmed that uploading confidential documents into a public AI tool places information in the public domain, constituting both a breach of client confidentiality and a waiver of legal privilege.

What does this all mean for AI and legal privilege in New Zealand?

To date, the impact of AI use on legal privilege has not been properly considered by the New Zealand courts. That’s likely to change - but in the meantime there are a number of points to note:

The issues aren’t clear cut: The overseas cases show different judges reaching different conclusions - and in particular on whether using a public AI programme amounts to waiver. As a Starting point, this shows that there’s likely to be a degree of uncertainty as to how the New Zealand courts will approach these issues - and the outcomes are likely to be fact-dependent.
Having noted that, the general principles of privilege will apply: The overseas decisions show the relevant courts applying existing principles of privilege to resolve these issues. It is highly likely that the New Zealand courts will adopt the same approach, subject to any specific AI-related law reform.
AI programmes are not “legal advisors” for the purposes of the Evidence Act 2006: Legal advice privilege (sometimes called solicitor-client privilege) only arises where professional legal services are sought or obtained for a legal advisor. AI programmes will not fall within this definition, so any ‘advice’ obtained from an AI programme - and, importantly, any information or material given to it for the purposes of obtaining legal advice - is at risk.
Confidentiality remains a key consideration: For other forms of privilege, confidentiality is a key consideration. If material isn’t confidential and treated as such, it is unlikely to attract legal advice privilege. Importantly, there’s a risk that providing material to an AI programme will fall foul of this - particularly in respect of AI programmes that are publicly available and/or those that allow for the collection, use and sharing of inputs in their terms and conditions. Licensed AI tools that integrate with an organisation’s systems and protect confidentiality offer greater protection against waiver arguments.

Time willime will tell - but care is necessary in the meantime

The New Zealand courts' approach to AI and privilege is still up for debate. Overseas cases shed light on the types of issues that arise, and how those issues might be approached. Ultimately though (and as these cases illustrate), outcomes are likely to be fact-specific - depending on the underlying circumstances, what AI programme is used, how it is used, and what is done with the output.

Until we have greater clarity, sensitive material should be approached with caution - particularly where disputes are likely. Cavalier use of AI may inadvertently waive privilege, which could in turn dramatically impact litigation outcomes and important legal rights.