Waiving goodbye to confidentiality: the risk of using open-source AI in investigations

A recent Upper Tribunal decision is the first in the English courts to comment on whether uploading confidential and privileged material to open-source AI tools[1] risks waiving legal privilege and breaching confidentiality.

In UK v Secretary of State for the Home Department [2026] UKUT 81, the Tribunal said that “uploading confidential documents into an open-source AI tool, such as ChatGPT, is to place this information on the internet in the public domain, and thus to breach client confidentiality and waive legal privilege”. The Upper Tribunal drew a distinction between open-source AI tools and closed-source AI tools, stating that closed-source AI tools which do not place information in the public domain, such as Microsoft Copilot, do not present this risk.

Whilst there may be an argument that any waiver of privilege from the use of an open-source AI tool is limited to a loss in relation to the AI tool provider, this has not yet been considered by the courts, and a cautious approach should be adopted.

In the context of workplace investigations, AI tools are increasingly being used to take notes and summarise interviews, review documents for relevant information, and finalise investigation reports. If the investigator uses an open-source AI tool for these purposes, there is a risk that confidentiality may be comprised which could be of significance if there are follow-on civil proceedings. Where participants in the investigation choose to feed sensitive information or documents which are relevant to the investigation into an open-source AI tool, this also risks breaching confidentiality.

Employers should note the following:

• The organisation should put in place an AI usage policy which includes a ‘no-paste’ rule for putting privileged or confidential information into open-source AI tools.
• The investigator should receive adequate training before they are permitted to use AI tools.
• If an open-source AI tool is to be used, any confidential or privileged information should be removed before prompts are submitted.
• Any AI outputs should be thoroughly reviewed to ensure that the information is accurate and that there are no hallucinations.
• Investigation participants, particularly the complainant, should be made aware of the risks of uploading documents and information to open-source AI tools and the impact this may have on any subsequent proceedings.
• While AI may assist with the administrative side of the investigative process, it cannot substitute for human decision-making.

If you would like to discuss any of the issues raised in this article, please contact Ed Henderson or Reshma Derasari.