Document AI-11-03 illustrates a refreshed approach to the draft reference document AI-09-06/Rev.3, based on outcomes from the 10th IWG on AI session (AI-10-04). The document addresses AI uses and potential risks in regulated automotive safety systems. It identifies AI-related risks across lifecycle stages—specification, model training, data management, verification, and operation—and describes risk management practices drawn from standards including ISO 26262, ISO 21448, ISO/PAS 8800, and UN R155/R156. Annex 1 provides illustrative AI use cases in perception, planning, motion control, driver assessment, and vehicle monitoring. Annex 3 tabulates illustrative practices for managing identified risks. The document establishes foundational concepts and terminology to support manufacturers, suppliers, and regulatory authorities in addressing safety implications of AI in automotive systems.

This document provides a draft form for submitting AI reference document change requests. Each request requires a separate submission. The form collects information including the date, agenda item, submitter affiliation, a title of twenty-five words or less, the location of the impacted section or paragraph number, original text, revised text, rationale, and discussion notes. The form includes fields for secretariat use to record whether requests are accepted, accepted as amended, deferred for further discussion, not accepted, or withdrawn.

This draft reference document consolidates input from documents AI-07-06, AI-09-03, AI-09-04, AI-08-07, and AI-09-05 to fulfill the initial deliverable of the IWG on Artificial Intelligence pursuant to WP.29/1190. The document addresses AI uses and benefits in automotive, establishes AI use cases prioritizing regulated automotive safety applications, identifies potential risks across AI lifecycle stages, and presents illustrative risk management practices informed by literature review. Seven whole-system lifecycle risks are identified, including insufficient documentation and traceability, inadequate organizational governance, and process drift. Additional risks span AI and use case specification, model architecture and training, data management, verification and validation, and in-use monitoring. Risk management practices derive from standards including ISO 26262, ISO 21448, ISO 23894, and UN Regulations 155 and 156, complemented by research and policy frameworks.