Legalithm logo
Chapter XIII — Final ProvisionsArticle 109

Article 109: Amendment to Regulation (EU) 2019/2144

Applies from 2 Aug 20268 min readEUR-Lex verified Apr 2026

Article 109 amends Regulation (EU) 2019/2144 (type-approval requirements for the general safety of motor vehicles and the protection of vehicle occupants and vulnerable road users) to integrate AI Act requirements. This Regulation mandates specific advanced driver assistance systems (ADAS) and automated driving features in new vehicles — including intelligent speed assistance, driver drowsiness and attention warning, advanced emergency braking, emergency lane-keeping, and event data recorders. Where these mandated systems use AI, they must now also comply with the AI Act. Always verify on EUR-Lex.

Start free assessmentAll articles

Who does this apply to?

  • -Vehicle safety system manufacturers developing AI-powered ADAS, automated driving, and occupant/VRU protection systems
  • -Type-approval authorities assessing compliance with mandatory vehicle safety features
  • -Automated driving system developers building systems that must meet the specific safety requirements of Regulation 2019/2144

Scenarios

A Tier 1 supplier develops an AI-based advanced emergency braking system (AEBS) for passenger cars that uses deep learning to detect and classify pedestrians, cyclists, and other vehicles from camera and radar input, triggering autonomous emergency braking when a collision is imminent.

Under Article 109, the type-approval for AEBS must address AI Act requirements. The AI-based perception and decision system is a safety-critical component: risk management (Article 9) must cover detection accuracy across lighting, weather, and road user types; data governance (Article 10) must ensure training data represents diverse pedestrian demographics, cyclist behaviours, and edge-case scenarios (children, wheelchair users, cargo bikes); and human oversight (Article 14) must define the driver's role in the braking decision chain. All documentation is integrated into the Regulation 2019/2144 type-approval technical file.
Ref. Art. 109, Art. 8, Art. 9, Art. 10

An automotive OEM implements an AI-powered driver drowsiness and attention warning (DDAW) system that monitors the driver's eye movement, head position, and steering behaviour using cabin cameras and ML-based analysis.

The DDAW system is mandatory under Regulation 2019/2144 for new vehicle types. Where the system uses AI for drowsiness/distraction detection, Article 109 requires AI Act compliance. Key concerns include: bias in detection algorithms across different demographics (Article 10 — training data must represent diverse face shapes, skin tones, and eyewear), alert threshold calibration (risk management must address both false alarms and missed detections), and transparency to the driver about how the system assesses their attention state (Article 13).
Ref. Art. 109, Art. 10, Art. 13, Art. 14

What Article 109 does (in plain terms)

Article 109 targets Regulation (EU) 2019/2144, which is the vehicle general safety Regulation — a landmark piece of EU legislation that mandates specific safety technologies in all new motor vehicles sold in the EU. Unlike Regulation 2018/858 (the framework type-approval Regulation amended by Article 107), Regulation 2019/2144 specifies concrete mandatory safety features:

  • Intelligent speed assistance (ISA) — AI-powered speed limit recognition
  • Driver drowsiness and attention warning (DDAW) — AI-based cabin monitoring
  • Advanced emergency braking (AEBS) — AI perception for collision avoidance
  • Emergency lane-keeping system (ELKS) — AI-based lane departure intervention
  • Event data recorders (EDR) — potentially AI-enhanced crash data capture
  • Automated driving systems — the Regulation lays the framework for automated driving type-approval requirements
  • Vulnerable road user (VRU) detection — AI-based pedestrian and cyclist detection, especially for trucks (direct vision requirements)

Many of these mandated features inherently rely on AI/ML for perception, classification, and decision-making. Article 109 ensures that the AI components of these specific mandatory safety systems are assessed against AI Act requirements as part of the type-approval process.

Read Article 109 together with Article 107 (which amends the framework type-approval Regulation 2018/858) — they are complementary, addressing different layers of the vehicle regulatory framework.

How Article 109 connects to the rest of the Act

  • Article 8 — Integration with sector legislation: AI Act requirements are assessed through the existing vehicle type-approval process. Article 109 activates this for the specific safety features mandated by Regulation 2019/2144.
  • Article 107 — Motor vehicle type-approval (Regulation 2018/858): the framework type-approval amendment. Articles 107 and 109 work in tandem — Article 107 for the framework, Article 109 for the specific safety feature requirements.
  • Annex I — Union harmonisation legislation: Regulation 2019/2144 is listed, confirming that its AI-equipped safety systems are subject to the integrated conformity assessment regime.
  • Article 10 — Data governance: critical for AI perception systems — training data must be representative of all road users, lighting conditions, weather, and geographic environments.
  • Article 14 — Human oversight: defines the driver's role alongside automated safety interventions — especially important for systems like AEBS that can override driver inputs.
  • Article 113 — Entry into force and application dates: confirms the timeline.

Practical guidance for vehicle safety system developers

For ADAS/safety system developers: - Map each mandatory safety feature under Regulation 2019/2144 to its AI components. For example, AEBS relies on AI-based object detection and classification; DDAW relies on AI-based driver state monitoring; ISA may rely on AI-based speed sign recognition. - For each AI component, prepare documentation addressing Articles 9–14 of the AI Act and integrate this into the type-approval technical file. - Pay particular attention to data governance for perception AI (Article 10): training datasets for pedestrian detection, drowsiness monitoring, and speed sign recognition must be representative across demographics, geographies, and environmental conditions. Document efforts to address bias and edge cases.

For OEMs integrating supplier systems: - Require Tier 1 suppliers to provide AI Act compliance evidence for each AI-powered safety component. Establish clear contractual allocation of compliance responsibilities. - Ensure that system integration does not introduce new AI risks not covered by the component-level assessment — e.g., sensor fusion that creates new failure modes.

For type-approval authorities: - Develop assessment procedures that evaluate AI Act compliance for each mandated safety feature specified in Regulation 2019/2144. - Coordinate with UNECE working groups (WP.29) on aligning AI Act requirements with international vehicle safety standards.

Compliance checklist

  • Map every AI component in Regulation 2019/2144 mandatory safety features: ISA, DDAW, AEBS, ELKS, VRU detection, automated driving systems, and event data recorders.
  • Classify each AI component under the AI Act risk framework — these are safety-critical systems that will qualify as high-risk under Annex I.
  • Ensure training data for AI perception systems is demographically, geographically, and environmentally representative (Article 10) — document bias assessment and mitigation.
  • Implement risk management (Article 9) addressing AI-specific failure modes: misclassification of road users, false activations, missed detections, and performance degradation in adverse conditions.
  • Design human oversight (Article 14) appropriate to each system: AEBS must allow driver awareness of system status, DDAW must have calibrated alert thresholds, ISA must support driver override.
  • Integrate all AI Act documentation into the Regulation 2019/2144 type-approval technical file.
Read the official text on EUR-Lex

Assess your vehicle safety AI compliance exposure — start the free assessment.

Start Free Assessment

Frequently asked questions

What is the difference between Article 107 and Article 109 for automotive AI?

Article 107 amends the framework type-approval Regulation (2018/858), which governs the overall process for approving vehicles and placing them on the EU market. Article 109 amends the vehicle general safety Regulation (2019/2144), which mandates specific safety features (AEBS, DDAW, ISA, etc.). They are complementary: Article 107 provides the framework-level integration, while Article 109 targets the AI in specific mandatory safety technologies.

Are event data recorders (EDRs) considered AI under Article 109?

A basic EDR that records predefined parameters (speed, braking, seatbelt status) using deterministic logic is unlikely to meet the AI system definition in Article 3(1). However, if an EDR uses machine learning to detect and classify crash events, reconstruct scenarios, or predict injury severity, the AI component would be subject to Article 109. Assess each system against the Article 3(1) definition.

How should bias in pedestrian detection AI be addressed under Article 109?

Article 10 (data governance) requires that training data be relevant, sufficiently representative, and free from errors. For pedestrian detection AI, this means the training dataset must include diverse demographics (age, ethnicity, body types, mobility aids), lighting conditions (day, night, dusk, glare), weather conditions (rain, snow, fog), and road user behaviours (jaywalking, group crossing, children). Document the dataset composition, any known underrepresentation, and the mitigation measures applied.