The organization should establish a clear process for defining and documenting all relevant requirements for newly developed AI systems or significant upgrades to existing ones. This includes functional, non-functional, technical, security, and ethical requirements.
The process should involve relevant stakeholders and incorporate considerations such as the intended purpose, potential risks, and impacts of the AI system. Documented requirements should serve as a foundational basis for the design, development, testing, and deployment phases of the AI system's lifecycle.