3.1 Disclosure of AI use and manipulated content
Individuals shall be clearly informed when they are interacting with an AI system or when artificial intelligence materially influences a process that affects them. AI-generated or materially manipulated text, images, audio or video shall be clearly identified where their artificial origin is relevant to authenticity, public understanding, legal proof or accountability, and appropriate provenance information shall be preserved. Disclosure shall be timely, prominent and expressed in language appropriate to the audience.
3.2 Transparency of purpose, role and scope
An organization shall clearly define and communicate why an AI system is used, the task it performs, the decisions or activities it may influence, and the boundaries within which it is authorized to operate. System-level transparency shall describe the system’s architecture, capabilities, limitations, intended users, and material dependencies sufficiently to enable responsible persons to understand and govern its role. A material expansion of the system’s functions, users, or influence over decisions shall require renewed approval and shall not occur through informal use or gradual function creep.
3.3 Decision explainability
Organizations shall ensure that persons who use or rely upon an AI system can understand why a particular decision, recommendation or outcome was produced and independently assess whether it should be accepted. The acceptable accuracy and speed of the explanation shall be proportionate to the importance and urgency of the decision. For consequential decisions, an adequate and timely explanation shall form an integral part of the decision-making process.
3.4 Decision traceability
Material AI-assisted decisions shall be traceable so that the process and outcome can be reconstructed. Relevant records shall identify the system and model version, material inputs and instructions, output, human intervention or override, and the final action taken.