We are all told that there has been a significant increase in the use of generative Artificial Intelligence (AI) tools by the legal profession. What is less often discussed is the use of AI by clients of law firms in relation to documents that are confidential and protected by legal professional privilege (privilege). While there has been no definitive ruling by a UK court on this question, it is important to be aware that sharing privileged information with an AI tool risks undermining the confidentiality of the materials in question and, in turn, may result in a loss of their privileged status.
What is privilege?
Back to the basics: what makes information or documentation privileged? Legal professional privilege (whether in the form of legal advice privilege or litigation privilege) depends on the information in question being confidential. In other words, if you post a legal advice note from your solicitor on social media, you cannot claim that it is privileged, as it has been shared with a number of other people and its confidentiality has been lost.
Problems with the use of AI tools
For all practical purposes, AI tools (especially those generally available to the public) should be treated in the same way as social media when it comes to the sharing of information. Most AI tools are provided, managed, and operated by third parties. When confidential information is input into an AI system, that information may be transmitted to external servers, stored, logged, or used for model training. Even where providers claim not to train models on user data, there is a risk, as terms of service often permit data retention for monitoring, security, or product improvement.
Once information is obtained by a third party who is not part of the solicitor–client relationship or subject to stringent confidentiality obligations, the confidentiality of that information is lost. The loss of confidentiality generally leads to a loss of privilege, which could result in privileged material having to be disclosed in litigation or under other disclosure obligations (for example, freedom of information requests or data subject access requests).
A recent lawsuit in the US state of California demonstrates that even the simplest AI tools used daily in modern office culture risk the loss of confidentiality. In Brewer v Otter.ai Inc, No 5:25-cv-06911 ND Cal, the court faced an allegation that Otter, in providing software that transcribes meetings on video-conference platforms, records these meetings without explicit permission from non-account holders. It is alleged that Otter then uses the data to train its speech recognition and machine learning models.
In a situation where a client is on a video conference with their solicitor and software of this type is used, it is clear that the confidentiality of the meeting comes under question and, consequently, so too does any privilege that might have attached to the information or legal advice shared during the meeting.
Similar issues arise if a client decides to 'double-check' advice provided by their solicitor and resorts to AI tools. For example, if a client ran a share purchase agreement (which contained a confidentiality clause) through an AI tool seeking 'alternative' solutions to their legal problems, the AI tool may provide a number of options. The client may share these with their solicitor, who would spend time, and therefore incur costs for the client, to ascertain whether the 'solutions' proposed by the AI tool were correct and not, for example, based on a clause which the AI had hallucinated, and which did not exist in the agreement.
In this scenario, not only would the client have breached the confidentiality provisions of the agreement, but they would also incur costs arising from the AI's hallucinations. Further, any documents or advice provided by the solicitor to the client would have lost their privileged status if processed through the tool.
What can be done?
It is, of course, unrealistic to expect clients to stop using AI tools altogether. However, they should exercise caution in any such use, particularly with information or documents received from their solicitors. Examples of this include legal advice notes, draft expert reports, and draft witness statements.
To minimise any risk of waiver of privilege, clients should consider:
- Avoiding input of sensitive or privileged information into public or consumer-grade AI platforms. There is very little clarity about how the information input into these platforms is used by the provider, and it is safe to assume that confidentiality is lost once the information is disclosed;
- Using enterprise or closed-system AI tools that contractually guarantee no data retention or model training. The terms and conditions under which these are provided should be carefully examined before proceeding;
- Implementing internal AI policies and training, ensuring staff members understand what can and cannot be shared; and
- In cases of highly confidential material, carrying out document risk assessments, demonstrating reasonable steps to preserve confidentiality if privilege is later challenged.
There is, of course, the practical question of proving that a client has used AI tools, which could have waived the privileged status of any information shared. This should offer little comfort, as it is likely that parties will seek court orders for disclosure of whether AI tools have been used to process particular information or documentation. The answer to such a question would not be privileged, and if a client is forced to answer in the affirmative, the other party might be able to establish a waiver in respect of actually privileged material.
The development and use of AI tools have undoubtedly raised novel questions around legal professional privilege and when it might be lost. Clients should be aware of the problems that indiscriminate use of AI tools can cause when it comes to their confidential materials. The law in this area has always been complex, and the advance of AI technology is likely to complicate matters further. If unsure of the consequences of using AI tools in relation to sensitive documents, we suggest seeking early legal advice to avoid potential pitfalls.
