# Vertical Governance Policy: Automotive — ISO 26262 / ISO 21448 (SOTIF) / UL 4600 # Doctrine v6 | R3 Adversarial Receipts # Last revised: 2025-07 schema_version: "1.0.0" vertical: automotive regime: ISO-26262/ISO-21448/UL-4600 effective_date: "2025-07-01" jurisdiction: Global-UNECE-WP29 meta: title: "Automotive AI Governance Policy — ISO 26262/SOTIF/UL 4600 Alignment" description: > Maps ISO 26262 Functional Safety, ISO 21448 (SOTIF), and UL 4600 Standard for Safety for the Evaluation of Autonomous Products to Doctrine v6 Λ-axes for AI/ML systems in automotive safety-critical applications (ASIL A–D). authority: "ISO 26262:2018 (all parts); ISO 21448:2022; UL 4600 Ed. 2 (2023); SAE J3016_202104; UNECE WP.29/2021/59" receipt_chain_required: true merkle_root_algorithm: SHA3-256 asil_level: ASIL-D regulatory_clauses: - clause_id: ISO26262-4-7 title: "Technical Safety Requirements — Specification" citation: "ISO 26262-4:2018 § 7; ISO 26262-6:2018 § 6" full_ref: "ISO 26262-4:2018 § 7 — Technical safety concept; ISO 26262-6:2018 § 6 SW safety requirements" lambda_axes: - axis: Λ5 label: Safety weight: 1.0 enforcement: mandatory rationale: > ASIL-D AI components must have verified safety goals with target probability of failure < 10^-8/h; safety case committed as receipt. - axis: Λ8 label: Robustness weight: 0.95 enforcement: mandatory - clause_id: ISO26262-6-9 title: "Software Unit Verification" citation: "ISO 26262-6:2018 § 9" full_ref: "ISO 26262-6:2018 § 9 — Software unit design and implementation (including ML per TR 4804)" lambda_axes: - axis: Λ7 label: Auditability weight: 0.95 enforcement: mandatory rationale: > All AI model versions and training data provenance must be versioned in the part management system; receipts link ASIL decomposition. - axis: Λ9 label: Explainability weight: 0.85 enforcement: mandatory rationale: > ML model inference paths for ASIL-B+ must be interpretable; saliency maps or decision trees provided as explanation receipts. - clause_id: ISO21448-8 title: "SOTIF — Evaluation of Triggering Conditions" citation: "ISO 21448:2022 § 8" full_ref: "ISO 21448:2022 § 8 — Identification and evaluation of triggering conditions causing hazardous behaviors" lambda_axes: - axis: Λ5 label: Safety weight: 0.98 enforcement: mandatory rationale: > AI perception systems must enumerate ODD (Operational Design Domain) corner cases; each triggering condition documented in receipt-linked hazard catalog. - axis: Λ4 label: Fairness weight: 0.70 enforcement: recommended rationale: > Pedestrian detection must demonstrate equitable performance across demographic groups (skin tone, age, mobility device); bias receipts required. - clause_id: UL4600-14 title: "Argument — Safety Case Sufficiency" citation: "UL 4600 Ed. 2 § 14" full_ref: "UL 4600 Ed. 2 § 14 — Safety case arguments: sufficiency, soundness, independence" lambda_axes: - axis: Λ1 label: Transparency weight: 0.88 enforcement: mandatory rationale: > Safety case must be public-facing (Transparency Index ≥ 0.8 per Doctrine v6 §2.1); receipt chain provides verifiable evidence ledger. - axis: Λ2 label: Accountability weight: 0.90 enforcement: mandatory - clause_id: SAE-J3016-L4 title: "Taxonomy of Driving Automation — Level 4 ADS" citation: "SAE J3016_202104 § 3.14" full_ref: "SAE J3016 Rev. Apr 2021 § 3.14 — Level 4: High Driving Automation ADS" lambda_axes: - axis: Λ2 label: Accountability weight: 0.95 enforcement: mandatory rationale: > Accountability for Level 4 decisions transfers to ADS; each decision receipt must include scene context hash and fallback state indicator. - axis: Λ10 label: Sovereignty weight: 0.80 enforcement: mandatory - clause_id: UNECE-WP29-R157 title: "Automated Lane Keeping Systems — Cybersecurity" citation: "UNECE WP.29 R 157 (2021); UNECE R 155" full_ref: "UNECE Regulation No. 155 — Cybersecurity and Cybersecurity Management System; R 157 Annex 4" lambda_axes: - axis: Λ6 label: Security weight: 0.92 enforcement: mandatory rationale: > CSMS must cover AI attack surfaces (model poisoning, adversarial inputs); threat analysis (TARA) results committed as security receipts. - axis: Λ8 label: Robustness weight: 0.88 enforcement: mandatory - clause_id: ISO26262-2-7-SGAS title: "Safety Goals — Automotive Safety Integrity Level Assignment" citation: "ISO 26262-2:2018 § 7; ISO 26262-3:2018 § 7" full_ref: "ISO 26262-3:2018 § 7 — Hazard analysis and risk assessment (HARA) with ASIL determination" lambda_axes: - axis: Λ5 label: Safety weight: 1.0 enforcement: mandatory - axis: Λ9 label: Explainability weight: 0.80 enforcement: mandatory rationale: > HARA risk classification for AI hazard sources must be explainable to type-approval authorities; decision rationale in receipt metadata. - clause_id: ISO-TR-4804 title: "Road Vehicles — Safety and Cybersecurity for ADS" citation: "ISO TR 4804:2020" full_ref: "ISO TR 4804:2020 — Road vehicles: Safety and cybersecurity for automated driving systems — Design, verification and validation" lambda_axes: - axis: Λ6 label: Security weight: 0.85 enforcement: mandatory - axis: Λ3 label: Privacy weight: 0.78 enforcement: mandatory rationale: > In-vehicle AI telemetry must comply with GDPR/ePrivacy; trip data used for model training must be consent-receipted. compliance_thresholds: minimum_lambda_coverage: 8 mandatory_axes: [Λ5, Λ6, Λ7] receipt_retention_days: 7300 # 20 years automotive liability max_perception_latency_ms: 50 asil_level: D odd_coverage_percent: 99.5 receipt_chain: algorithm: SHA3-256 chaining: merkle_dag quorum: 2-of-3 nodes: [primary, redundant-ecu, cloud-archive] hardware_security_module: true