Article 106: Amendment to Directive (EU) 2016/797
Article 106 amends Directive (EU) 2016/797 (on the interoperability of the rail system within the European Union) to integrate AI Act requirements. AI systems in railway interoperability components — such as AI-based signalling (ETCS/ERTMS), traffic management, predictive maintenance, and autonomous train operation — must comply with the AI Act alongside the railway interoperability rules. Conformity assessment and EC verification of interoperability constituents and subsystems must now address AI-specific obligations. Always verify on EUR-Lex.
Who does this apply to?
- -Railway system component manufacturers using AI in signalling, control-command, rolling stock, or infrastructure subsystems
- -National safety authorities (NSAs) and the EU Agency for Railways (ERA) overseeing interoperability and safety authorisation
- -Rail operators and infrastructure managers deploying AI-based traffic management, predictive maintenance, or automated train operation systems
Scenarios
A signalling company develops an AI-enhanced ETCS (European Train Control System) on-board unit that uses machine learning to improve braking curve calculations based on real-time track conditions, train load, and weather data, going beyond the standard deterministic braking model.
A rail infrastructure manager deploys an AI-powered predictive maintenance system that analyses sensor data from track, switches, and overhead contact lines to forecast failures and schedule preventive maintenance before disruptions occur.
What Article 106 does (in plain terms)
Article 106 integrates the AI Act into Directive (EU) 2016/797, the Railway Interoperability Directive. This Directive ensures that the EU rail system functions as a single interoperable network by setting requirements for subsystems (structural and functional elements of the rail system) and interoperability constituents (components incorporated into subsystems).
The amendment means that where AI systems are embedded in railway interoperability constituents or subsystems — whether in control-command and signalling (e.g., ETCS/ERTMS), rolling stock, infrastructure, energy, or traffic management — the EC verification and conformity assessment must account for AI Act obligations. This follows the Article 8 integration model.
The railway sector is rapidly adopting AI across the value chain: automated train operation (ATO, from Grade of Automation 1 to 4), AI-enhanced signalling, predictive maintenance, passenger flow optimisation, and energy management. Article 106 ensures these systems are not assessed solely under traditional railway safety frameworks but also meet the AI Act's requirements for risk management, transparency, and human oversight.
How Article 106 connects to the rest of the Act
- Article 8 — Integration with sector legislation: AI Act requirements for high-risk systems are assessed through existing product-level conformity assessment for Annex I products. Article 106 brings railway interoperability into this framework.
- Annex I — Union harmonisation legislation: after the amendment, railway interoperability constituents and subsystems with AI fall under the integrated AI Act assessment regime.
- Article 9 — Risk management: railway AI systems must address sector-specific safety risks, including the Common Safety Method for risk evaluation (CSM-RA) under Regulation (EU) 402/2013, integrated with the AI Act's risk management requirements.
- Article 14 — Human oversight: critical for automated train operation — train drivers or remote operators must retain the ability to intervene, consistent with the applicable Grade of Automation.
- Article 113 — Entry into force and application dates: confirms when Article 106 becomes enforceable.
Practical guidance for railway stakeholders
For signalling and component manufacturers: - Identify all AI components in your railway products: ML-based braking curve optimisation, AI-enhanced object detection for ATO, adaptive traffic management algorithms, and predictive component health monitoring. - Prepare AI Act compliance documentation and integrate it into the technical file for EC declarations of conformity (interoperability constituents) and EC verification (subsystems). - Engage your Designated Body (DeBo) or Notified Body (NoBo) early to discuss how they will assess AI components within the existing railway conformity assessment framework.
For rail operators and infrastructure managers: - Review specifications for new AI-equipped railway systems to require AI Act compliance evidence from suppliers. - For ATO deployments, ensure the human oversight framework (Article 14) aligns with the operational concept and Grade of Automation — from driver advisory systems (GoA 1) to unattended operation (GoA 4). - Integrate AI system monitoring into existing safety management systems (SMS) required under Directive (EU) 2016/798.
For ERA and national safety authorities: - Consider how AI Act requirements interact with existing CSM-RA (Common Safety Method for Risk Assessment) obligations — the AI Act's risk management requirements should complement, not duplicate, the railway safety risk assessment. - Develop or endorse guidance on AI assessment competence requirements for NoBos and DeBos in the railway sector.
Compliance checklist
- Identify all AI systems embedded in railway interoperability constituents and subsystems subject to Directive (EU) 2016/797.
- Classify each AI system under the AI Act risk framework — safety-critical AI in railway control-command, signalling, or rolling stock will generally qualify as high-risk under Annex I.
- Integrate AI Act documentation (risk management, data governance, technical documentation, human oversight) into the EC verification and conformity assessment technical files.
- Ensure the risk management approach for AI components is consistent with both the AI Act (Article 9) and the railway Common Safety Method for Risk Assessment (CSM-RA).
- For automated train operation (ATO) systems, define human oversight levels appropriate to the Grade of Automation and document compliance with Article 14.
- Engage NoBos/DeBos early to confirm their readiness and methodology for assessing AI components in railway subsystems.
- Verify application dates under [**Article 113**](/en/ai-act-guide/article-113) and the consolidated text on [**EUR-Lex**](https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=OJ:L_202401689#article-106).
Check if your railway AI system needs compliance action — start the free assessment.
Start Free AssessmentRelated Articles
Frequently asked questions
Does Article 106 apply to metro and light rail systems?
Directive 2016/797 covers the EU rail system as defined by the Directive, which primarily targets the interoperable main-line rail network. Some Member States may extend its application to metro and light rail through national legislation. If a metro or light rail system falls under Directive 2016/797 in your Member State, Article 106 applies to its AI components. Check national transposition measures.
How does the AI Act interact with the railway CSM-RA (Common Safety Method for Risk Assessment)?
The CSM-RA (Regulation (EU) 402/2013) requires systematic risk evaluation for significant changes to the railway system. The AI Act adds AI-specific risk management requirements (Article 9). In practice, manufacturers should integrate AI risk assessment into their CSM-RA process — addressing AI-specific hazards (data quality, model drift, adversarial inputs) alongside traditional railway safety hazards. The two frameworks are complementary.
What about AI in non-safety railway applications like passenger information?
If the AI system is not part of an interoperability constituent or subsystem assessed under Directive 2016/797, Article 106 does not apply directly. However, the AI system may still be subject to the AI Act through other classification routes (e.g., Annex III high-risk areas, or transparency obligations under Article 50 for AI systems that interact with passengers).