AUSTRALIA Trends and Developments Contributed by: John Lee, Michael Williams, Siabon Seet and Vanessa Farago-Diener, Gilbert + Tobin
including generative AI (GenAI) tools such as ChatGPT, poses an interesting challenge for the protection of trade secrets and other confiden - tial information given AI’s ability to analyse, learn and decipher complex patterns and data sets. Typically, the use of public AI platforms will involve the user sending their query to the third- party provider of the platform (likely located out - side Australia) via the internet. This begs several questions: does this “disclosure” result in any confidential information contained within the query losing its confidential character? Does it depend on the applicable terms of use? What can the provider do with the information received via queries to its AI tool? Looking at the elements of the equitable action for breach of confidence, it would present a sig - nificant challenge for a plaintiff trying to argue that the information within a query submitted to a publicly available AI tool, was received by the provider of that tool in circumstances import - ing an obligation of confidence (unless perhaps there are particular settings or other options enabling that designation). Further, disclosure in those circumstances may well demonstrate an absence of sufficient care on the plaintiff’s part in limiting access to the information outside the organisation, which adds to the difficulty in establishing the continuing status of the infor - mation as secret or confidential. On the other hand, for an organisation using an enterprise version of a third-party AI tool rather than a version freely open to the public, the terms of the agreement between the organisation and the AI provider are likely to address confidential - ity, security, and access issues. Depending on the terms, this may provide a better basis on which to argue that the AI provider must keep user inputs confidential, and more broadly, that
the information uploaded into the tool maintains its confidential status. Much will depend on the facts of the case, including the tool being used and how it is pro - vided (eg, whether it is open or closed), as well as the terms (if any) between the provider and the user (or the user’s employer). Certainly, it is crucial for organisations to develop appropriate guidance and policies for employees and con - tractors regarding the use of AI tools, to avoid inappropriate use and potentially negatively con - sequences. Users of AI tools must be conscious not only as to whether the information they are inputting into an AI tool is confidential information of the organisation but also whether it is third-party confidential information that has been disclosed under a non-disclosure agreement or other confidentiality provision. Inputting confidential information of a third party into an AI tool (and thereby disclosing that information to the rele - vant vendor) could potentially cause the organi - sation to be in breach of its contractual obliga - tions (dependent on the terms of the underlying contractual provisions). Another important element for organisations to be aware of is the ability of third-party AI pro - viders to use information submitted by users as training data for the tool. This heightens the risk from a confidentiality perspective, particu - larly in light of real-life instances where certain GenAI tools appear to have memorised aspects of their training data and are able to regenerate those inputs in near-identical form (potentially as outputs for other users). The safest option is to ensure AI developers are prohibited from train - ing their models on user prompts, including via appropriate contractual prohibitions where pos - sible, and using opt-out facilities when available.
36
CHAMBERS.COM
Powered by FlippingBook