This article is part 2 of our new series about law in the AI age. In this article, we explore how users of generative AI platforms like ChatGPT cannot use legal professional privilege to prevent disclosure in Court of their online discussions. Part 1 of our series, about legal liability for AI advice, is available here.
We live in an age where generative AI large language models (LLMs), including the wildly popular Chat GPT, have become mainstream. Users in Australia and all over the world are increasingly turning to these platforms for advice about all manner of personal things, from romantic relationships to health tips.
It comes as no surprise that advice about legal matters is also a popular conversation piece, with ChatGPT confirming when prompted that:
Legal queries are among the top professional categories of ChatGPT usage, alongside programming and business strategy.
When conversations about legal problems are conducted with a qualified lawyer in Australia, the client is typically protected by the doctrine of ‘legal professional privilege’ (LPP). This body of law allows clients to refuse to disclose to the Court confidential, and potentially damaging, documents which have the dominant purpose of legal advice. Yet, for the reasons discussed below, interactions with LLMs are not protected from disclosure to Courts. Users of LLMs need be aware that legal communications with these gen-AI platforms are able to be fully laid bare in litigation.
The Australian doctrine of LPP has the aspiration of protecting the administration of justice and the right of individuals and other entities or organisations to obtain confidential advice about their legal circumstances.
The ‘privilege’ belongs to the client, meaning that your lawyer may only disclose privileged information to the Court if instructed to do so by you as the client.
The Full Federal Court in Pratt Holdings Pty Ltd v Commissioner of Taxation [2004] FCAFC 122 held that communications between a solicitor and client will be privileged if it can be established that they were prepared for the purpose of giving or obtaining legal advice, regardless of whether the third party is acting as an agent and if litigation is on foot or reasonably necessary.
In Grant v Downs (1976) 135 CLR 674, the majority judgment of the High Court held that:
“…a document which was produced or brought into existence either with the dominant purpose of its author, or of the person or authority under whose direction, whether particular or general, it was produced or brought into existence, of using it or its contents in order to obtain legal advice or to conduct or aid in the conduct of litigation, at the time of its production in reasonable prospect, should be privileged and excluded from inspection“.
In using a generative AI platform like ChatGPT, a user may have “…the dominant purpose of…using it or its contents in order to obtain legal advice or to conduct or aid in the conduct of litigation.”
For example, here is an extract of the start of a real chat with ChatGPT:
User: I think we breached our contract, can you help me work out if we are in the wrong?
Chat GPT: I can help you assess whether a breach has occurred and what the potential consequences might be. To give you an accurate view, I’ll need a few details….Can you share the relevant parts of the contract (or the whole contract if you prefer)…..What exactly happened?
These communications relate to matters that are actually or potentially litigious in nature, and ordinarily they would be protected from disclosure under the doctrine of LLP where a solicitor-client relationship arises. Yet LPP only applies to communications between a client and their lawyer acting in a professional capacity.
As we discussed in our recent article on taking legal advice from AI, LLMs are not lawyers, are not admitted to legal practice, and are not bound by professional obligations or ethical duties such as confidentiality under the Legal Profession Uniform Law or equivalent state legislation.
Courts interpret LPP narrowly, and Australian authority is clear that privilege is an important, but limited, legal doctrine that only applies when all criteria are met, including the communication being made by a legally qualified advisor.[1]
This aligns with the position under sections 118 and 119 of the Evidence Act 1995 (NSW), that LPP exists in relation to communications or documents where there is a “client and lawyer relationship”. As an LLM is not human and cannot be a lawyer, the communication is not protected from disclosure to the Court.
Under the Uniform Civil Procedure Rules 2005 (NSW), parties to proceedings are required to disclose documents that are directly relevant to the issues in dispute. For the purpose of discovery, the term ‘document’ is broadly defined and could include electronically generated materials such as prompt histories, draft outputs, or notes produced with the assistance of an LLM.
LLM interactions do not attract LPP protection. Accordingly, any input entered into, or output generated by, an LLM that is relevant to the proceedings may be discoverable, including drafts of pleadings, affidavits, or legal arguments that were informed or produced with the assistance of an LLM.
Generative AI models cannot hold a legal practising certificate or give qualified legal advice. Discussions that users have with these models about legal problems do not and cannot attract the protection of LPP. Users must be aware that the content of these interactions, whilst seeming to provide instant and largely free legal help, can be fully laid bare before the Court, and seriously prejudice their legal position.
To talk to a qualified legal professional about your legal issues, please contact Edwards + Co via our contact details below.
[1] See various cases, including Esso Australia Resources Ltd v Commissioner of Taxation (1999) 201 CLR 49.
