Generative AI tools and English litigation: what might have to be disclosed?

Generative AI tools are now embedded in the workflows of many organisations. Staff can chat with those tools as if they were talking to a human being, with their prompts and the tools’ responses being retained on the organisation’s systems.

Helpful as these tools may be, most people will not have front of mind when they are using them that their “conversations” (prompts inputted and responses received) may comprise disclosable documents for the purposes of any subsequent English litigation - and that they may be unhelpful. This is particularly so because these tools encourage informal, stream-of-consciousness back-and-forth, and may also embolden staff to seek advice on areas where they might ordinarily consult a professional, such as a lawyer.

Why might prompts and responses be disclosable?

It is fundamental to English litigation that the parties must disclose to each other any documents within their control which are relevant to the issues in the case.

The term “document” is very broadly defined and includes anything in which information of any description is recorded.

The definition is certainly broad enough to capture prompts to an AI tool, and the tool’s responses to those prompts, if they are relevant to the issues in a case.

Prompts and responses can be very revealing. We have already seen a case, albeit in a slightly different context, where a fee earner at a law firm was required to disclose their informal exchanges with an AI tool to the court, and where those exchanges were then set out in full in the public judgment.

Can prompts and responses be protected by privilege?

A question then arises as to whether prompts and responses can be protected by legal professional privilege.

If a document is protected by privilege, it can be withheld from disclosure to third parties - including the court and the other side in litigation.

There are two main types of privilege that can apply:

1. legal advice privilege in broad terms protects confidential communications between client and lawyer made for the dominant purpose of obtaining legal advice;
   
   
2. litigation privilege in broad terms protects confidential communications between client and lawyer, client and third party, or lawyer and third party, made for the dominant purpose of litigation which is either already underway or in the client’s reasonable contemplation.

Legal advice privilege

Anecdotally, it is becoming common for people to seek legal advice from AI tools. However, these tools are not lawyers and so English courts may take the view that prompts and responses in this context are not communications between a client and lawyer. This means they would not attract legal advice privilege and would likely be disclosable in any relevant litigation. 

The door may be ajar to argue otherwise following a recent decision, Aabar Holdings v Glencore. This suggests for the first time that internal client communications made for the purpose of obtaining legal advice are protected by legal advice privilege, even if no lawyer is involved in them (including, for example, emails between individuals within a corporate exchanging thoughts in preparation for a meeting with a lawyer). One can see an argument that preparatory communications exchanged with an AI tool prior to speaking with a lawyer may fall into the same bucket. However, at this early stage, legal advice privilege protection for prompts and responses between an individual and an AI tool is far from guaranteed, even if those prompts and responses on their face concern legal issues.

Litigation privilege

Litigation privilege extends beyond communications between clients and their lawyers to communications between clients and third parties, or between lawyers and third parties. It would therefore appear that “conversations” with an AI tool can be protected by litigation privilege - on the basis that they are essentially on the same footing as a communication between a client and a third party - provided that all the other elements of the litigation privilege test are met.  

Confidentiality

For a communication to be protected by privilege of either type, it needs to be confidential. Although not necessarily the last word on this point, there is already an English case which suggests that communications with public AI tools are not confidential, as their contents are used to train the tool and could be reverse engineered by others or appear in responses to others. This concern does not apply to enterprise tools with confidentiality protections in place. 

Practical pointers

• Avoid using public AI tools for business-related matters.
• Consider use of enterprise AI tools carefully and ensure appropriate policies are in place to guide staff on how to use them.
• Ensure staff are aware that their “conversations” with (prompts to and responses from) an AI tool may potentially be disclosable in any relevant litigation.
• For sensitive matters, and in particular those where legal advice may need to be sought or which may involve litigation, consider seeking specific advice in advance about how AI tools should be used, and how any prompts to them should be framed, to maximise the possibility of privilege protection.

Our team are on hand to support you with any actual or potential dispute, whether in the form of a litigation, arbitration or investigation - please get in touch with any of our team to discuss the issues raised above or any other needs.

Related News, Insights & Events