Self-driving cars are no longer a purely future phenomenon. Vehicles with far reaching automated functions are already on Dutch roads, while experiments with vehicles without a driver present in the vehicle are legally possible under conditions through article 149aa paragraph 1 of the Road Traffic Act. This inevitably raises the question of who is criminally liable when such a vehicle causes an accident resulting in injury or death. The short answer is that Dutch criminal law still starts from human or organisational addressees of the norm, such as the driver, the operator, or the legal entity behind the system. An autonomous system is not a criminal subject in its own right.
The classic traffic offences as a starting point
The starting point lies with the well known traffic offences of the 1994 Road Traffic Act. Article 6 criminalises a traffic accident caused by fault that results in death or serious bodily injury. Article 5 and, since 2020, article 5a paragraph 1 function as lighter and heavier endangerment norms respectively, even without actual injury occurring. The criminal seriousness is further shaped by articles 175, 176, 178 and 179, including the possibility of an additional driving disqualification.
The standard for establishing fault under article 6 is strict and nuanced. The District Court of Limburg held that it cannot be stated in general terms whether a single traffic violation is sufficient for fault, because the assessment must look at the conduct as a whole, the nature and concrete seriousness of the violation, and the other circumstances of the case. A recent ruling by the District Court of Gelderland illustrates this sharply. A single driving error, consisting of insufficient control over the vehicle, amounted to endangerment under article 5 but not to the culpable carelessness required under article 6.
For automated vehicles this distinction is essential. If a human driver relies too heavily on the system, ignores warnings, or uses the vehicle outside its permitted operational domain, criminal liability remains readily conceivable. But if the system itself makes an error without the human supervisor reasonably having had to or been able to intervene, applying article 6 becomes considerably more difficult.
What current Dutch practice already shows
The type approval granted by the RDW in April 2026 for Tesla’s FSD Supervised system illustrates how the legislator and the regulator currently frame this relationship. The RDW explicitly emphasises that a vehicle with FSD Supervised is not self driving. It is an advanced driver assistance system that can take over many driving tasks, but the driver remains responsible and must be able to intervene at all times. Sensors monitor whether the driver keeps their eyes on the road and their hands available to take over the wheel immediately. If alertness is insufficient, warning signals follow, and the system can ultimately be temporarily disabled.
This example confirms the line taken in case law. As long as a system is legally and technically approved as a driver assistance system rather than a fully autonomous system, the driver remains the addressee of the norms in articles 5 and 6. The type approval therefore changes nothing about the starting point of human responsibility, while at the same time making clear that the technology in practice already goes well beyond ordinary cruise control, which does make the factual assessment of fault and culpability in an individual case more complex.
The experimental regime for driverless vehicles
A separate regime applies to vehicles without a driver present in the vehicle itself. Article 149aa paragraph 1 requires a permit for such experiments. That permit may include exemptions from various statutory provisions, but explicitly not from articles 5 and 6. Even in experimental settings, the core criminal traffic norms therefore remain fully in force. Article 149ab paragraph 1 further requires that the permit sets out which safety measures are taken, how monitoring and evaluation take place, who the driver is and where they are located, and how many vehicles that driver may control at the same time. This is not merely administrative technicality, but can also provide important evidence for criminal attribution after an incident.
Liability of the legal entity
When a fault cannot convincingly be attributed to a human road user, the legal entity comes into view. Article 51 paragraph 1 of the Criminal Code explicitly allows for the criminal prosecution of legal entities, as well as prosecution of the principal or the person who actually directed the prohibited conduct. In the context of self-driving vehicles this can be relevant for manufacturers, software companies, or operators, for example when structurally defective software updates are rolled out, known safety warnings are ignored, or vehicles are knowingly put into traffic while known safety problems exist. The mere fact that software makes an error is not sufficient. The case will have to be traced back to concrete, culpable organisational choices, such as inadequate testing procedures, neglected updates, or instructions that led to the prohibited conduct.
Causation as a separate obstacle
With autonomous or semi-autonomous vehicles, several possible causes often coincide, such as a human supervision error, a software error, a sensor malfunction, defective map data, or the conduct of other road users. A violation of a norm and a criminally relevant outcome do not automatically coincide. Illustrative is a ruling by the District Court of North Holland, which held that a Tesla that had come to a stop in time had thereby broken the causal chain. In future cases concerning automated driving as well, it will always have to be separately established that the identified fault, and not some other factor, caused the accident and its consequences.
Evidentiary problems concerning technical data
Where classic traffic cases often turn on speed, visibility, distance and reaction time, automated driving adds software versions, event data recorder data, vehicle logs, sensor data and update history. The European Court of Human Rights recognises that electronic and technical sources of evidence can raise particular questions of authenticity and integrity, and that rejecting defence evidence in technically complex cases must be properly reasoned in light of article 6 of the Convention. This can become relevant in this context when the defence requests access to source data, vehicle telemetry, or an independent expert in algorithmic decision making.
The limit set by article 7 of the Convention
The principle of legality under article 7 of the European Convention on Human Rights sets a clear limit on any creative expansion of criminal liability. Criminal liability must be based on accessible and foreseeable legislation and case law. Judicial clarification of existing norms is permitted, but an arbitrary or unforeseeable expansion of them is not. For autonomous vehicles this means that the absence of technology specific legislation does not mean prosecution is never possible, but it does mean that liability must be capable of being grounded in existing, sufficiently clear norms. Criminal law cannot construct a culprit merely out of social dissatisfaction with a serious outcome.
Frequently asked questions
Is the driver always criminally liable in an accident involving a self-driving car?
No. The driver in principle remains the addressee of the norms in articles 5 and 6 of the Road Traffic Act, but whether he is actually criminally liable depends on whether he can be blamed. If he used the system correctly, within its permitted operational domain, and did not reasonably have to intervene, culpable carelessness within the meaning of article 6 is difficult to establish.
Can a vehicle manufacturer be criminally prosecuted for an accident?
In theory yes, on the basis of article 51 of the Criminal Code, but only when the accident can be traced back to a concrete, culpable organisational choice, such as an inadequate testing procedure, ignoring known safety warnings, or rolling out a defective software update. The mere fact that software makes an error is not sufficient for the manufacturer’s criminal liability.
Does the type approval of Tesla FSD Supervised mean the car is self driving?
No. The RDW has explicitly confirmed that FSD Supervised is an advanced driver assistance system and not a self driving system. The driver remains responsible, must keep their eyes on the road, and must be able to intervene immediately. This also has criminal law consequences, because the driver thereby remains the addressee of the norms in articles 5 and 6.
Is it already permitted to drive fully driverless cars in the Netherlands?
Only within the framework of experiments for which a permit has been granted under article 149aa. Even then, articles 5 and 6 remain fully applicable, and the permit must set out, among other things, who the driver is, where they are located, and which safety measures apply.
What role does vehicle data such as logs and sensor data play in a criminal case?
An important and growing role. In automated driving, it is often necessary to rely on software versions, event data recorder data, sensor data, and update history to establish exactly what happened and whether there is a causal link to the accident. This kind of technical evidence raises its own questions about reliability and accessibility for the defence.
Can someone be held criminally liable purely because the law does not yet contain specific rules for self-driving cars?
No. The principle of legality under article 7 of the European Convention on Human Rights does not allow liability to be stretched merely because an outcome feels socially unsatisfactory. Liability must always be capable of being based on existing, sufficiently clear norms.
Conclusion
The Netherlands currently has a workable, though not technology specific, criminal law framework for accidents involving automated vehicles. That framework still focuses on human and organisational culpability, not on the autonomous system as an independent perpetrator. The type approval of Tesla FSD Supervised shows that the regulator, too, holds to this line, by explicitly recording that the driver remains responsible. In future cases, three questions will therefore be decisive. Who was the addressee of the norm, which concrete fault is culpable, and can the accident be convincingly attributed to that fault both causally and evidentially. For manufacturers, operators and software developers it is therefore important to properly document testing procedures, update processes and internal responsibilities, so that after an incident it is clear which choices were made and by whom. Do you have questions about liability in relation to vehicle automation, or are you involved in a dispute in which this issue plays a role? Please feel free to contact Law & More for a no obligation conversation.
