Short answer
EU law does not assign every automated-vehicle accident to a single “AI owner.” Several regimes can apply at once. National traffic and tort law allocates civil fault or strict vehicle liability. Compulsory motor insurance protects accident victims. EU product-liability law can make manufacturers and other economic operators liable for defective hardware, software, updates or connected services. Vehicle type-approval law regulates whether the automated driving system may enter service. The AI Act, Data Act, GDPR and cybersecurity rules govern system design, monitoring, data and incident response, but breach of those rules is not automatically the complete civil-liability answer.
Responsibility follows the failed function and the party controlling it. Preserve vehicle, sensor, event-recorder, software-version, update, remote-command, maintenance and security evidence. Determine whether the driver was expected to supervise, whether the automated driving system stayed inside its operational design domain, whether a safe fallback occurred and whether later software or services remained under manufacturer control.
The applicable dates matter. The revised Product Liability Directive applies to products placed on the market or put into service after 9 December 2026. A July 2026 AI Act amendment moved the main high-risk duties for systems embedded in regulated products to 2 August 2028, while the Data Act already applies and the Cyber Resilience Act remains phased.
Begin with the accident, not the technology label
Reconstruct the dynamic driving task: who or what steered, braked, perceived objects, planned the path and issued warnings at each moment. Record the automation level, operational design domain, handover design, driver state, road conditions, connectivity, maps, cybersecurity alerts and any remote intervention.
The contractual chain may include the vehicle manufacturer, automated-driving-system developer, component and sensor suppliers, software or map provider, importer, distributor, fleet operator, owner, driver, repairer and remote-operation service. More than one can be liable, with contribution allocated later.
Traffic liability and motor insurance remain the victim-facing route
Member States retain important rules on driver fault, keeper or owner liability, causation, damages and procedure. Automation changes the evidence but does not displace those national rules. A driver may be responsible for ignoring a lawful takeover request; a fleet operator for unsafe deployment or maintenance; and a manufacturer for a defect that made the request impossible to follow.
, as amended by , requires civil-liability insurance for covered vehicle use and supports direct victim claims, compensation bodies and cross-border protection. Insurers can compensate first and pursue responsible manufacturers or operators under applicable recourse rules.
The insurance framework covers vehicle use “consistent with the vehicle's function as a means of transport.” —
The Court's earlier judgment in Vnuk, , treated use consistent with a vehicle's normal function broadly. The amended definition now provides the current textual anchor. Whether a particular automated movement falls within national compulsory cover still requires the Directive and implementing law.
Product liability reaches software and post-sale control
For products already on the market, the national law implementing Directive 85/374/EEC remains relevant. modernises the regime for products placed on the market or put into service after 9 December 2026.
“A product shall be considered defective where it does not provide the safety that a person is entitled to expect.” —
“Product” includes software. Components include intangible items and related services integrated or interconnected with the product. Potential defendants include manufacturers of the vehicle and defective component, importers, authorised representatives, fulfilment providers and, in defined circumstances, distributors or online platforms. A person substantially modifying a vehicle can be treated as manufacturer of the modified product.
Defect assessment can consider design, instructions, reasonably foreseeable use, connectivity, learning capacity, cybersecurity, recalls and applicable safety requirements. Manufacturer control can continue after sale through authorised updates, upgrades, machine-learning behaviour and related services. Failure to provide necessary safety updates can prevent a post-market defect defence.
The claimant normally proves damage, defect and causation, but disclosure and presumptions address technical complexity. Boston Scientific, , under the older directive, illustrates that safety expectations and product-group evidence can matter; its medical-device holding should not be mechanically applied to vehicles.
Type approval defines the safety baseline
adds general safety and automated-vehicle requirements within the EU type-approval system. sets procedures and technical specifications for type-approval of automated driving systems in fully automated vehicles, including system safety, operational design domain, human-machine interaction, data recording, cybersecurity interaction and in-service reporting.
Approval is not immunity from civil liability. A system can comply at approval yet become unsafe through production variation, deployment outside its domain, deficient updates or a newly discovered vulnerability. Conversely, an accident does not alone prove regulatory non-compliance or defect.
The AI Act is a compliance layer with phased dates
, the AI Act, treats an AI safety component subject to listed product-safety legislation as high-risk when third-party conformity assessment is required. For vehicle systems, the interaction with sector type approval and implementing legislation must be mapped carefully.
High-risk duties cover risk management, data governance, technical documentation, logging, transparency, human oversight, accuracy, robustness, cybersecurity, quality systems and post-market monitoring. Earlier provisions already apply in phases, but the main high-risk-system timetable was extended in July 2026. Verify the specific role and date before alleging breach.
, in force since 27 July 2026, narrowed and clarified the concept of a safety component and changed the timetable. Chapter III, Sections 1 to 3 now apply from 2 December 2027 for Annex III systems and from 2 August 2028 for Article 6(1) product systems. It also permits targeted limitation of overlapping AI Act requirements where listed sector legislation provides equivalent protection. For an automated vehicle, classification therefore depends on the system's intended safety function and the applicable vehicle-approval route; integration into a regulated product is not enough by itself.
The AI Act does not create a universal compensation action. Its records and standards may nevertheless inform defect, negligence, causation and regulatory enforcement.
Data access, privacy and cybersecurity shape the evidence
The applies from 12 September 2025. Users of connected products can obtain readily available product and related-service data and request transfer to third parties, subject to trade-secret, security and personal-data safeguards. It can support repair, diagnostics and aftermarket services, but it is not an unrestricted accident-discovery code.
The GDPR still requires a legal basis, purpose limitation, minimisation, security and respect for data-subject rights. Vehicle users may not be the only data subjects; passengers and bystanders can be captured.
The applies principally from 11 December 2027, with reporting obligations applying from 11 September 2026 and conformity-assessment provisions from 11 June 2026. It excludes products covered by certain sector rules to the extent equivalent requirements apply, so vehicle components need a scope analysis. Product-liability and national negligence may still attach to insecure design or poor vulnerability handling.
Preserve and align the evidence
Issue preservation notices immediately. Secure event data recorder and ADS data, raw sensor data where available, logs, calibration, diagnostics, software bill of materials, model and map versions, update history, certificates, safety case, incident reports, service records, driver alerts, consent and access records. Preserve hashes and chain of custody.
Avoid overwriting evidence through routine updates or remote diagnostics. Give investigators proportionate access while protecting personal data, trade secrets and system security. Contractual indemnities should allocate investigation and recourse but cannot remove mandatory victim rights.
Conclusion
Automated-vehicle harm is resolved through layered responsibility, not a single AI rule. Motor insurance provides the immediate victim-facing framework; national law allocates traffic responsibility; product liability tests safety defects; type approval and the AI Act set compliance evidence; and data and cybersecurity rules govern connected operation. The decisive work is a time-specific reconstruction of control, defect, causation and every party's real role.
Sources
This article provides general legal information, not advice for a specific matter. Laws, facts and procedural deadlines may change the result. Follow the linked primary sources and obtain local advice before acting.