General guidance for drafting the baseline document on artificial intelligence in the automotive sector. The document emphasizes focusing revision on green text in AI-11-03, using change requests for substantive modifications processed during IWG plenary sessions, and integrating successful requests into a single master document. Four sections are addressed: Introduction (AAPC lead) to clarify purpose and scope; AI Uses & Benefits (OICA/CLEPA lead) to provide intuitive understanding and define terms; Risks (modifications by change request); and Risk Management (China and EME leads) to clarify operationalization in certification regimes while maintaining agreement-neutral, conceptual language at reference document level, avoiding prescriptive guidance.

Report of the 11th session of the IWG on AI held on 2 September 2026 online. The session discussed the status of the draft reference document, with the USA formally objecting to the refresh plan (AI-11-03, AI-10-04) and insisting that AI-09-06/Rev.3 presented to WP.29 was the only valid baseline document. The co-chairs and majority of experts, including the UK, Germany, France, Canada, the Netherlands, EU Commission, OICA/CLEPA, and SAE, supported using AI-11-03 as the baseline. The group decided to establish two small drafting groups to address green text in sections 2 and 4, with changes submitted through change request forms for review during plenary sessions.

The session covers adoption of the agenda and brief report of the 11th session, discussion of work plan for formal document submission to the June 2027 WP.29 session, discussion of draft reference document AI-11-03 including modifications and change requests to sections 1 through 4 and annexes, presentations on the use of AI in the automotive sector, discussion of timeline and future sessions, and small drafting group discussions on 28 October.

Study on image generation method to support evaluation of automated driving systems. Fine-tuned an image generator on the Zenseact Open Dataset using 100,000 road-scene images annotated with 3D bounding boxes. Demonstrated ability to generate plausible images reflecting scenario specifications and to create images causing detection errors in object detectors. Next steps include quantitatively evaluating validity and realism of generated images and developing methods to automatically identify scenario specifications that degrade object detector performance.

Comments on the structure and content of the draft reference document on artificial intelligence in regulated automotive safety systems. Notes a six-month timeline for adoption by WP.29 in June 2027, with formal submissions due by March 2027. Recommends prioritizing a future-proof repository of information on potential AI-related risks and risk management that can be maintained through standard WP.29 processes. States current regulatory frameworks are sufficient to address unreasonable safety risks from AI use in vehicles. Suggests the document should include introductory explanatory text for non-specialists, clear definitions of technical terms, accessible risk summaries, and a developed risk-management section to support WP.29 regulatory activities.

Presentation proposing an AI Management System (AIMS) as a complementary organizational layer to existing safety, cybersecurity, and software update management systems. AIMS would regulate AI through organizational processes rather than technical requirements, covering data lifecycle, model development, verification and validation, change control, and operational monitoring. The system is technology-neutral, lifecycle-based, and aligned with industry standards including ISO 26262, ISO 21448, and ISO/IEC 42001, avoiding the need for product-specific AI test procedures and re-approval of model updates.

Proposal to modify the definition of Validation Data to replace ‘a final’ with ‘candidate’ and append clarification that it is used for model selection or tuning while independent test data used for final performance reporting are not used for model selection or tuning, and add a definition of Test Data as data used to evaluate the final selected model and not used for fitting, model selection or tuning, with a note on dataset separation to avoid leakage between sets.

Proposal to add text to para. 3.2.4.6 clarifying that uncertainty-related risks may propagate from upstream to downstream AI modules when uncertainty information is inadequately communicated or used, affecting the resulting action, and to add text to Annex 2, item 1.6 clarifying that inadequate communication or use of uncertainty information across AI modules can propagate or amplify uncertainty-related risks and affect the resulting action, even when the upstream estimate is informative.

Proposal to add text to para. 3.2.4.6 clarifying that uncertainty-related risks may propagate from upstream to downstream AI modules when uncertainty information is inadequately communicated or used, affecting the resulting action, and to add text to Annex 2, item 1.6 clarifying that inadequate communication or use of uncertainty information across AI modules can propagate or amplify uncertainty-related risks and affect the resulting action, even when the upstream estimate is informative.

Proposal to append sentences to paragraph 2 addressing uncertainty-related risks propagated from upstream to downstream AI modules and supporting uncertainty-aware action, and to paragraph 3 considering whether uncertainty information remains meaningful when used by downstream AI modules and how it affects the resulting action.

Proposal to add working definitions for uncertainty quantification, aleatoric uncertainty, epistemic uncertainty, and uncertainty-aware action at the end of section 1.3 of AI-11-03, Annex 5. These definitions do not replace definitions in applicable standards and support general discussion of uncertainty and its effects on action.

Proposal to modify Annex 1, 1.1.2 to replace the clause beginning ‘and such applications fall within the scope’ with language clarifying that applications fall within the first mode of AI use only where AI is part of the regulated product and directly performs or supports a safety-relevant function, and to add that AI tools used solely for development or assessment fall under section 2.3.3.

Proposal to modify the definition of concept drift in para. 3.2.9.2 and Annex 2, para. 6.2 to clarify that concept drift arises when changes alter the relationship between inputs and target outputs, and that a change in input distribution alone does not necessarily imply a change in this relationship.

Proposal to insert a sentence in para. 2.2.1 clarifying that decision trees can be learned from data and not all are manually specified rule systems, and to modify para. 2.2.5 to state that Generative AI models can use different architectures, with some being Transformer-based and potentially foundation or general-purpose AI models.

Proposal to modify para. 4.2.5.2 to replace ‘management of changes in data and operational distributions’ with ‘management of changes in data and process assurance’. Where a closed-loop data chain is applied to vehicle AI systems safety development activities, evidence of safety assurance capability of the complete closed-loop data chain shall be provided, including safety assurance descriptions for data collection, data processing, and data annotation processes. Where data training is used as an improvement measure for functional insufficiency, relevant analysis shall be conducted and the rationality of the data shall be justified to ensure the improved system meets system-level safety-related requirements.

Add one NB note beneath the “AI Safety” definition in Annex 5, Section 1.1, item 4, clarifying that the definition uses “safety” in the functional-safety sense and that hazards affecting AI-based safety functions may be initiated by malicious actions or arise from personal data processing, and that assurance of AI systems interacts with cybersecurity and data-privacy requirements such as UN R155 and ISO/SAE 21434.

Proposal to add principles and stages of risk-driven lifecycle approach to para. 4.2.1.1, replacing ‘risk informed lifecycle approaches explanation’ with ‘Risk-driven lifecycle approach explanation’ and adding explicit reference to principles of risk-driven lifecycle approach and stages of risk-driven lifecycle approach.

Proposal to add to illustrative practices in para. 4.2.6.3 a reference to scenario coverage assessment based on ISO 34502 and test case generation based on ISO 34505 as sources of evidence for AI system assurance, and to add ISO 34501-34505 to Annex 4 A standards list.

Change request to revise the driving function name from ‘Acceleration Control Pedal Error’ to ‘Acceleration Control for Pedal Error’ in Section 1.2.4, based on UNECE guidance that the original term is not a correct driving function name.