Contribution for the AI literature review.

Contribution for the AI literature review.

The document presents a non-exhaustive list of automotive use cases of AI aligned with WP.29/1182, incorporating contributions from AI-02-04 and AI-02-05. It focuses on use cases regarding the use of artificial intelligence in regulated automotive safety systems. The list is split between driving functions (perception, planning and motion-control) and non-driving functions (driver occupant monitoring systems and monitoring systems for vehicle components or functions), following a risk-based approach.

This document provides a consolidated list of contributions from the IWG on AI experts to a literature review activity on Artificial Intelligence. References are organized by type and listed alphabetically, including standards such as ISO 21448:2022, ISO 23894, ISO 26262, ISO 22989, ISO 39003:2023, ISO/IEC 42001:2023, ISO/IEC TR 5469, ISO/PAS 8800:2024, ISO/SAE 21434:2021, ISO/TS 5083:2025, SAE J3016, SAE J3298, SAE J3312, and UL4600. Policy papers and research articles address trustworthy AI, autonomous vehicles, and safety.

At the third IWG on AI session, experts agreed that a robust review supporting tasks specified in ECE/TRANS/WP.29/1186, para. 6 was needed as a precursor to drafting a reference document. The IWG on AI agreed that three guiding questions provide the basis for common understanding of artificial intelligence in the automotive sector, including emergent practices, risks, and how AI is currently being used to promote safety. The three questions address where and how AI is used and its benefits, what risks AI presents, and what emergent practices exist for AI use and risk mitigation.

This document contributes to consolidated text and a risk catalogue for artificial intelligence in the automotive sector. It defines AI as a field of computer science focused on creating systems capable of performing tasks requiring human intelligence, including learning, reasoning, problem-solving, natural language processing, and decision-making. High-priority AI use cases include perception, planning, and motion-control driving functions, as well as driver assessment systems. The document identifies 14 risks organized across five lifecycle stages: AI and use case specification, model architecture and training processes, data specification and management, verification and validation, and operation and in-use monitoring. It provides detailed risk descriptions and references emergent practices for managing these risks, including ISO/PAS 8800, NIST AI RMF, and EU AIA approaches. Lower-priority use cases referenced include AI-06-09 for periodic technical inspection and AI-06-12 for cybersecurity threat detection.

Proposal to add a new section on Cross-Cutting Organisational and Assurance Risks to AI-07-06. Six risks are proposed: insufficient organisational governance for AI safety management; inadequate definition of roles, responsibilities, and competencies; lack of independence between development and verification activities; degradation of AI safety processes over time; ineffective change management for AI-related artefacts and processes; and inadequate management of AI supply-chain risks. Each risk addresses factors across the lifecycle that can influence the effectiveness of technical AI risk mitigations.

Consolidated comments on table 1 of AI-06-07 from AI-08-05, AI-08-06, AI-08-09, AI-08-11, AI-08-12, and AI-08-13. The document provides a prospective catalogue of risks related to the use of AI in the automotive sector, organized according to AI lifecycle stages: AI & Use Case Specification; AI Model Architecture & Training Processes; Data Specification & Management; Verification & Validation; and Operation/In-Use Monitoring. For each identified risk, high-level and sub-level descriptions, brief risk descriptions, and potential management and mitigation approaches are provided.

This consolidated draft reference document, based on WP.29-195-20, addresses Artificial Intelligence in the automotive sector. It defines AI-based systems as connectionist systems trained using machine learning algorithms. The document identifies AI use cases for driving functions including perception, planning, motion control, and end-to-end systems, plus non-driving functions like driver assessment. It establishes prospective risks across five AI lifecycle stages: specification, model architecture and training, data management, verification and validation, and in-use monitoring. The document provides initial risk management approaches informed by literature review, noting that current regulatory provisions may require evaluation to address AI-specific testing and updating needs.

The Informal Working Group on Artificial Intelligence identified relevant literature for a literature review addressing existing standards, research, best practice, and regulations relating to artificial intelligence in automotive safety applications. Proposal to include IEEE standards on algorithmic bias, robustness testing, and organizational governance; risk assessment frameworks from IEEE and NIST; the EU AI Act; international instruments including the Council of Europe Framework Convention on AI and OECD AI Principles; and the G7 Hiroshima AI Process Code of Conduct. These sources address lifecycle phases and risk management relevant to the prospective catalogue of AI risks in automotive (AI-07-06) and consolidated draft documentation (AI-06-03).

The IWG on AI 7th session (March 12, 2026) adopted the agenda and 6th session report. The IWG secretariat presented updated AI use case document (AI-07-03) and consolidated literature review (AI-07-04). France presented additions to use cases (AI-07-13); China presented literature review proposals (AI-07-07, AI-07-08, AI-07-09, AI-07-10, AI-07-11, AI-07-12). The IWG expressed support for the guiding questions document (AI-07-05) and directed language amendments. Japan presented a draft reference document framework (AI-07-14); the prospective catalogue of risks (AI-07-06) received general support with requests for amendments addressing bias, risk management approaches, flexible non-exhaustive risk listing, and software update impacts. The secretariat was directed to compile a consolidated draft reference document for June WP.29 submission.

The IWG will be informed on the status of identification of AI Use Cases, the revised draft Terms & Definitions Document, and the ongoing collection of proposals for the literature review. The IWG may wish to exchange views concerning guiding questions for AI in automotive, risks of the use of AI in automotive and their management and mitigation, and the consolidated draft for reference document. The 9th IWG on AI session is planned for June 3-4th, hybrid in London, UK and online.

This document provides a prospective catalogue of risks related to AI use in the automotive sector, organized across five AI lifecycle stages: AI & Use Case Specification, AI Model Architecture & Training Processes, Data Specification & Management, Verification & Validation, and Operation/In-Use Monitoring. For each of 23 identified risks, the document lists relevant ISO/PAS 8800, NIST AI RMF, EU AIA, and other standards or frameworks that address potential management and mitigation approaches. Extended descriptions explain how risks such as blackbox behaviour, data poisoning, distribution shift, insufficient test coverage, and concept drift may compromise safety in automotive AI systems.

For consideration in section 3 of AI-09-06, consolidated draft/initial output for a reference document, proposal to add inspectability of AI-enabled regulated systems throughout operational life, including lifecycle compliance verification through PTI and in-service monitoring activities with standardized access to diagnostic information, software version verification, AI safety status indicators, event logs, sensor calibration status, ADAS calibration verification, and system health monitoring functions, and to add use of AI by inspection authorities including AI-assisted brake condition analysis, tyre wear analysis, emissions anomaly detection, automated lighting inspection, vehicle structural anomaly detection, computer vision for exterior and underbody inspection, and aggregated analysis of in-service defect patterns.

Proposal to insert recommended requirements for AI implementation in regulated automotive safety systems based on AI-08-08, AI-06-11, and AI-04-09 documents. The recommended requirements address identification and classification of risk levels, derivation processes for general and AI-specific requirements, requirements regarding datasets, definition of test criteria and thresholds, iterative evaluation processes, and continuous processes throughout vehicle lifetime. Safety measures for AI systems must be implemented with severity appropriate to potential risks, with gap analysis conducted against design specifications to identify AI-specific risks not covered by existing standards.

This document updates terminology and definitions for the WP.29 Informal Working Group on Artificial Intelligence, amending AI-01-02/rev.2 as presented at the IWG’s second session. It compiles technical terms for AI in regulated automotive safety systems across four sections: Fundamental AI Concepts & Terminology, AI Methods & Models, AI Characteristics, and Data-Related Concepts & Terminology. Definitions draw from ECE/TRANS/WP.29/1182 and ECE/TRANS/WP.29/1186, among other sources, and are arranged as a living reference document supporting harmonization of AI practices for road vehicles. The document introduces new terms including Convolutional Neural Network, Neural Network, and Transformer, alongside established definitions such as Machine Learning, Generative AI, and Trustworthiness.

This document provides a consolidated list of contributions by the IWG on AI experts to the literature review activity. References are organized by type: Standards including ASPICE 4.0, IEEE standards on robustness testing, algorithmic bias, and organizational governance of AI, ISO standards on safety of intended functionality, functional safety, AI concepts and terminology, ethical considerations, AI management systems, functional safety and AI systems, road vehicle safety and AI, cybersecurity engineering, and safety for automated driving systems, SAE standards on levels of driving automation, AI data, and AI use cases, and UL4600; Policy Papers and Research Articles addressing testing autonomous vehicles, trustworthy AI, ethics of connected and automated vehicles, and safety effectiveness; Terminology Databases including IEC Electropedia and ISO Online Browsing Platform; and Regulations including Council of Europe Framework Convention on AI, EU AI Act, UN Regulation No. 155 on cybersecurity, and UN Regulation No. 156 on software updates.

The document presents a consolidated list of artificial intelligence use cases in regulated automotive safety systems, aligning with ECE/TRANS/WP.29/1182 and incorporating input from IWG on AI sessions 2–8, including contributions AI-02-04, AI-02-05, AI-06-06, AI-09-03, and AI-09-06. High priority driving function use cases include perception, planning and motion-control systems. Non-driving function use cases include driver and occupant assessment systems, and vehicle component monitoring systems. Lower priority use cases address AI in periodic technical inspection and cybersecurity, as discussed in sessions referencing AI-06-09 and AI-06-12.

The 8th session of the IWG on AI convened on 24 April 2026. Experts discussed AI use cases, including support for dividing driver and occupant assessment systems and objection to inclusion of adaptive front lighting systems. The chair directed production of draft text on connections between AI risks and use cases. The secretariat was directed to reinstate definitions for transformer architecture and convolutional neural network. Experts reviewed literature, guiding questions, and AI risks and management practices. The chair directed production of an updated consolidated draft with tables for AI risks and illustrative practices for AI risk management. The 9th session is scheduled for 3–4 June 2026 in London.

France supports retaining risks 3.5 (unreliable uncertainty estimation) and 6.1 (adversarial input manipulation) in Table 1. France favours secretarial text for mitigations 3.1 (blackbox behaviour), 3.2 (lack of robustness), and 3.4 (model over/underfitting) in Table 2. In Annex I, para. 11 addressing lack of robustness, France favours deleting the industry proposal on explanation of mitigations, pending merger with the mitigations table subject to level of detail determination.

This document presents AAPC comments on AI-09-06, a consolidated reference document on artificial intelligence in regulated automotive safety systems. The proposed structure consolidates AI uses, use cases, and risk management into single chapters supported by annexes containing tables and bibliographies. The document should support future regulatory deliberations without pre-empting regulatory requirements, presenting factual statements in neutral language. Considerations should address risks not captured by conventional testing. Examples to illustrate system-specific aspects that impact whether the application of AI presents new concerns and determines the nature of responses include predictive window defogging, predictive vehicle maintenance, and AI vehicle knowledge systems.

The IWG on AI will hold its 9th session on June 3–4, 2026 in London. The agenda includes adoption of the agenda and brief report of the 8th session (AI-09-01), consideration of AI use cases, literature review and terms & definitions proposals, guiding questions for WP.29 submission, and discussion of the consolidated draft reference document. The IWG will receive presentations from stakeholders on AI use in the automotive sector and discuss future meetings and work direction beyond the initial mandate.

Comments on the draft consolidated reference document for AI use in regulated automotive safety systems. The comments propose editorial changes to the introduction; minor editorial modifications; and substantive contributions to AI use cases in driving functions (section 3.2) and non-driving functions (section 3.3). Canada’s contributions include additions to risk descriptions such as lack of resilience, regression or catastrophic forgetting, bias as systematic performance disparities, and degradation of AI safety processes over time; expanded guidance on risk management practices with specific automotive examples; and clarification of verification, validation, data management, and operational monitoring requirements aligned with existing standards including ISO 26262, ISO 21448, ISO 23894, and UN Regulations 155 and 156.

Proposals for revisions to Section I including highlighted deletions, addition of two paragraphs (2.3, 2.4) in Section II with a request to add safety benefits to para. 2.6, editorial requests for text placement in Section III introducing use case content, small editorial proposals in Section IV’s introduction, merging of whole lifecycle and organizational risks, removal of para. 4.6 (systematic performance disparities) in favour of para. 4.5, revision of para. 6.2 (concept drift), removal or moving of para. 6.5 to whole lifecycle risks, and substantive suggestions in Section V for deletion, merging and restructuring of Table 2 with standard and title references only. Annex 1 critiques the need for extended descriptions.

This initial consolidate draft reference document for artificial intelligence in regulated automotive safety systems, based on WP.29-195-20, AI-07-08, AI-07-11, AI-07-09, AI-07-10, and AI-07-12, establishes AI use cases and identifies potential risks across the AI lifecycle. The document addresses driving functions including perception, planning and motion control, and end-to-end systems, as well as non-driving functions such as driver assessment and occupant monitoring. It presents six major AI lifecycle risk categories: whole system risks including insufficient documentation and governance; specification risks such as improper definition of intended use; model and training risks including blackbox behaviour and lack of robustness; data management risks such as data poisoning and distribution shift; verification and validation risks including insufficient test coverage; and operational risks including concept drift and limited failure detection. Illustrative risk management practices drawn from standards, research, and regulations are provided for each identified risk to support manufacturers, suppliers, and approval authorities.

The USA expert submitted initial red-line comments on the consolidated reference document for artificial intelligence in regulated automotive safety systems, marking areas falling outside the Informal Working Group on AI’s terms of reference and bracketing sections pending internal consultations. The USA indicated it will provide additional technical comments following stakeholder consultations.

AI-Enabled Road Vehicles Managing a Complex Standards Landscape presents the current landscape of standards and regulations for AI in road vehicles, structured overview of core AI governance, engineering, organisational and data governance standards required for EU AI Act conformity and UNECE compliance, and identifies gaps and challenges in alignment. ISO 8800 (2029) Road vehicles Safety of Artificial Intelligence is proposed as a common framework providing navigation and connection across horizontal AI standards with automotive sector-specific needs, reducing fragmentation and enabling transparent, traceable evidence-based assurance for regulators, industry and society while supporting compliance with EU AI Act and future UNECE requirements.