ACE: AI Privacy Orchestrator

AI operation as a cloud service is proliferating in a variety of offerings, including AI-as-a-Service (AIaaS). The reliance of AI on data introduces significant privacy concerns, as trained models bear knowledge of the consumed data. This means privacy restrictions on input data traverse to the trained model, which then requires coordinated access and usage control, corresponding to the sensitivity levels of consumed data. This need is unmet yet by existing management software solutions, as they largely rely on manual configurations to achieve the desired accessibility control. However, manual management is proven to be prone to human errors, delays and high operational cost, particularly when the management user - in this case likely to be an AI developer - lacks the expertise in managing cloud applications.

Our innovation will give businesses, particularly those focused on data commercialisation and AI services but with limited expertise in cybersecurity and/or cloud service management, the ability to preserve the privacy of their data and AI services in a flexible manner. At the same time, enable better (re)utilisation of trained machine learning models in multiple clients' environments. ACE is a user-friendly software that provides AI developers a systematic methodology to define the privacy requirements of their AI services, and use the requirements to automatically coordinate the configurability of access and usage control to comply with said requirements.

To this end, ACE alleviates the burden of having to manually configure restrictions on AI services or manage their allocation to clients' cloud environments. Unlike existing technologies, through novel analytics and orchestration functions, ACE enables privacy-by-design deployment of AI services, without compromising privacy requirements. Our technology will enable privacy-preservant and secure (re)usability even of sensitive data, to ensure adequate training of machine learning models to provide AI services to digital ecosystems.