Automotive ML component — what evidence will an assessor expect?
Informational only, not professional or legal advice. Facts are linked to the source we checked; statuses can change after the date above.
The buyer question is “what evidence will the assessor want for the ML component?” No public page we could reach answers it with a checklist. What the public pages do give is a map of where each kind of evidence is expected to live.
Component and item level. ISO/PAS 8800:2024 is titled “Road vehicles — Safety and artificial intelligence”, and ISO 26262-1:2018 is the road-vehicle functional safety vocabulary. We could not retrieve the ISO catalog text in this review (the server returned HTTP 403), so we quote nothing from either page and do not summarize their clauses.
Function-level framing. The IEC catalog entry for ISO/IEC TR 5469:2024, “Artificial intelligence - Functional safety and AI systems”, says it describes “use of AI inside a safety related function to realize the functionality;” and “use of non-AI safety related functions to ensure safety for an AI controlled equipment;”. Those are two different evidence stories: arguing the model itself, or arguing the non-AI function around it.
Vehicle-level safety case. The UL 4600 product page is titled “Standard for Evaluation of Autonomous Products” and describes “building and evaluating a safety argument for vehicles that can operate in an autonomous mode.” It also refers to “the scope of the safety case.” The evidence an assessor sees here is an argument, not a single test report.
Regulatory layer in the EU. The AI Act Service Desk Annex I lists, among motor-vehicle legislation, “Regulation (EU) 2018/858 of the European Parliament and of the Council of 30 May 2018 on the approval and market surveillance” (the page text we read is cut off there) and “Regulation (EU) 2019/2144 of the European Parliament and of the Council of 27 November 2019 on type-approval requirements” (also cut off).
Editorial judgment: A practical evidence pack separates three layers: model-level evidence (data, training, test results), function-level evidence (the non-AI monitor or fallback and its integrity argument), and vehicle-level evidence (the safety case that ties them to the operating domain). We have not found an official public sentence stating which test results an approval authority or assessor accepts for a learned component; ask the assessor early and get the answer in writing.
Sources
- ISO/PAS 8800:2024 - Road vehicles — Safety and artificial intelligence - ISO
- ISO 26262-1:2018 - Road vehicles — Functional safety — Part 1: Vocabulary - ISO
- UL 4600 - Standard for Evaluation of Autonomous Products - UL Standards & Engagement
- ISO/IEC TR 5469:2024 (IEC Webstore) - IEC
- AI Act Service Desk: Annex I - European Commission