Article 107: Amendment to Regulation (EU) 2018/858

Article 107 is arguably the highest-impact amendment article in Chapter XIII, because the automotive sector represents the largest-scale deployment of AI in safety-critical products. It amends Regulation (EU) 2018/858, which is the framework regulation for EU type-approval and market surveillance of motor vehicles (M, N, O categories — passenger cars, commercial vehicles, and trailers) and their systems, components, and separate technical units.

The amendment integrates AI Act requirements into the vehicle type-approval process. When a motor vehicle or its systems contain AI — from advanced driver assistance (lane keeping, adaptive cruise control, automatic emergency braking) to fully autonomous driving systems — the type-approval authority must assess compliance with both Regulation 2018/858 and the AI Act. This follows the Article 8 integration model: a single, unified conformity assessment rather than parallel regulatory tracks.

Given the complexity of automotive AI (perception stacks using cameras, radar, and LIDAR; planning algorithms; driver monitoring; V2X communication) and the safety-critical nature of the domain, this amendment creates a substantial compliance obligation for the automotive industry.

• Article 8 — Integration with sector legislation: the master provision for folding AI Act requirements into existing product conformity assessment. Article 107 activates this for motor vehicles.
• Annex I — Union harmonisation legislation: Regulation 2018/858 is listed, meaning motor vehicle AI systems are assessed through the integrated regime.
• Article 9 — Risk management: automotive AI risk management must cover the operational design domain, edge cases, sensor failures, adversarial conditions, and the full range of driving scenarios.
• Article 10 — Data governance: autonomous driving perception systems are trained on massive datasets; data governance requirements (representativeness, bias management, data quality) are critical.
• Article 14 — Human oversight: defines how driver-vehicle interaction must be designed — driver handover protocols for Level 3, remote operation for Level 4, and monitoring interfaces.
• Article 109 — Vehicle safety (Regulation 2019/2144): closely related amendment covering general vehicle safety requirements, including ADAS and automated driving — read Articles 107 and 109 together.
• Article 113 — Entry into force and application dates: confirms the timeline for automotive AI compliance.

For OEMs: - Conduct an AI inventory across all vehicle platforms: perception systems (camera, radar, LIDAR-based AI), planning and decision-making algorithms, ADAS features, driver monitoring, voice assistants integrated with vehicle controls, and over-the-air (OTA) update systems. - For each AI system, create an AI Act compliance dossier integrated into the type-approval technical file. This must cover: risk management (Article 9), data governance (Article 10), technical documentation (Article 11), automatic logging (Article 12), transparency (Article 13), and human oversight (Article 14). - Establish a process for managing AI Act compliance through OTA updates — when a vehicle's AI is updated post-market, reassess whether the update constitutes a significant modification requiring reassessment. - Brief senior management on the dual-track obligation: vehicle type-approval and AI Act compliance are now a single integrated requirement.

For Tier 1/2 suppliers: - Provide OEM customers with comprehensive AI documentation packages that enable them to meet type-approval requirements. Include: training data descriptions, model architectures, performance benchmarks (precision, recall, latency), known limitations, and failure mode analyses. - Establish contractual clarity on who bears responsibility for AI Act compliance elements at the component vs. vehicle system level.

For type-approval authorities: - Develop assessment methodologies for AI components in motor vehicles, potentially leveraging UNECE WP.29 work on automated driving (e.g., UN Regulation No. 157 for ALKS) alongside AI Act requirements. - Invest in technical competence: AI assessment requires expertise in machine learning, computer vision, and safety engineering beyond traditional mechanical/electrical type-approval skills.

A unique challenge in the automotive sector is over-the-air (OTA) software updates. Modern vehicles can receive AI model updates, feature unlocks, and safety patches remotely. Under Article 107:

• OTA updates that modify the AI system's behaviour may constitute a significant modification, triggering reassessment under both Regulation 2018/858 and the AI Act.
• OEMs must establish processes to evaluate each OTA update against AI Act compliance criteria before deployment.
• Post-market monitoring (Article 72) must capture real-world performance data and feed it back into the risk management system — creating a continuous compliance loop rather than a one-time certification.
• Type-approval authorities may need to develop streamlined reassessment procedures for AI updates that do not fundamentally alter the system's safety profile.