Responsible AI Reference Architecture Overview
An effective Responsible AI Reference Architecture allows for the deployment of AI in enterprises to be carried out in a responsible manner, ensuring safety for users and the business, and security of the models and datasets. There is currently no recognised standard which specifies the safety and security services required for AI, however the ISO 42001: AI Management System and the key requirements in the EU AI Act provide some level of guidance.
By analysing these and other sources, a Responsible AI Reference Architecture can be created to identify and categorise the security services required to be in place for a complete safe and secure environment for AI. This provides both enterprise-wide governance services as well as solution-specific safety and security services. This is shown below.
The AI Reference Architecture identifies the required security services and organises them into the following categories: governance, assurance, risk management, identity & access management, datasets, AI and cyber defence, incident management and IT services.
The Responsible AI Reference Architecture provides a coherent view of the services required to effectively manage AI. By adopting it, enterprises can be assured they are well aligned with the requirements of the ISO 42001 AI Management System and are prepared for any external audit of their AI activities.