The application of privilege to generative AI tools — such as ChatGPT, Microsoft 365 Copilot and legal-specific platforms — is an emerging issue in Australian law, with key questions still developing around how privilege applies to both inputs and outputs of these tools. Two main questions arise: Are the communications with the tool privileged, and does inputting privileged material into a tool waive any privilege?

Assuming that the tool is only used for the purposes of a lawyer providing legal advice, confidentiality becomes the main issue, both for privilege arising and in relation to the risk of waiving already privileged material input. AI tools are typically not operated by the client or law firm, so care must be taken to ensure that confidentiality is maintained from the public and the vendor. This is not a new issue, as clients and firms have disclosed privileged material to third-party vendors for some time, such as cloud storage providers.

Privilege arising: Assuming that the client/lawyer's purpose in inputting material into the tool is a legal one (such as asking it to draft a letter of demand to be settled by a lawyer), if the communications back and forth with the tool are confidential, privilege should arise. This largely depends on whether the inputted content is used to train the AI model, which, if it is, typically has implications for confidentiality. Further, the contractual terms of the tool used will dictate the confidentiality regime, including whether the vendor has access to the relevant content.

Waiver: If a lawyer or client inputs privileged material into an AI tool, whether privilege is preserved depends on the confidentiality issues raised in the previous paragraph.

The Law Council of Australia has cautioned that improper use of generative AI tools may carry ethical, professional and legal risks, particularly where client confidentiality is not assured.

Australian courts are beginning to issue protocols and guidance on AI use in legal practice.