AI in Vehicles Safety and Reliability Must Be Ensured

Related Vendors

DEKRA Automobil GmbH

FAULHABER Drive Systems

Wuxi InfiMotion Technology Co., Ltd.

The Dekra Traffic Safety Report 2025 "Mobility in Changing Times" addresses, among other things, the integration of artificial intelligence (AI) in road vehicles. These technologies hold significant potential for increased road safety but also carry high risks. Read what Dekra (Global leader in vehicle inspections) experts are calling for here.

Artificial Intelligence (AI) is increasingly being integrated into road vehicles. This inherently holds significant potential for improved traffic safety. However, "the high risks associated with the—never entirely avoidable—failure of AI systems must not be overlooked," warns Xavier Valero, Director Artificial Intelligence & Advanced Analytics at Dekra. Therefore, regulations and standards, such as the EU AI Act as a legal framework for the development of such systems, are all the more important. The topic is one of many covered in the current Dekra Road Safety Report 2025 "Mobility Through the Ages."

Driver Assistance Systems Classified As High-Risk Systems

In the future, Advanced Driver Assistance Systems (ADAS) will take on important tasks in an increasing number of vehicles:

• For the detection of traffic signs, pedestrians, cyclists, or vehicles,
• for monitoring tire pressure or engine temperature, or
• for initiating automated braking or lane-changing maneuvers

However, since a malfunction could endanger human health and life, the EU classified such ADAS as high-risk systems in its AI Act passed in 2024. "These systems, which are crucial for driving safety, must therefore be subject to strict oversight and validation to ensure their reliability and avoid risks," emphasizes Xavier Valero.

A comprehensive quality management system throughout the entire AI lifecycle is required for full compliance, as it ensures that all activities remain traceable and verifiable.

Xavier Valero, Director Artificial Intelligence & Advanced Analytics at Dekra

AI Law Allows Exceptions

The AI Act also allows exceptions if AI systems have already been evaluated in accordance with existing regulations, such as vehicle type approval (Regulation EU 2018/858). This means that safety features like ADAS, while indirectly regulated by the AI Act, can be validated through existing type approval procedures without requiring additional third-party assessments.

"Aside from ensuring that their systems meet the requirements of the AI Act, system providers must also establish a comprehensive AI management system with clear guidelines and workflows within their organization," adds the Dekra expert. Additionally, all parties involved—upstream technology suppliers, downstream integrators, and manufacturers—must contribute to ensuring system safety, data security, and transparency. "A comprehensive quality management system throughout the entire AI lifecycle is required for full compliance, as this ensures that all activities remain traceable and verifiable," says Xavier Valero.

Continuous Monitoring During Operation

Guidelines for the safety of AI systems in the automotive industry are provided by ISO/PAS 8800. This standard covers the entire AI lifecycle, from design to deployment, ensuring that systems are both safe and reliable at every stage. It complements ISO 26262 (functional safety) and ISO 21448 (SOTIF—Safety of the Intended Functionality) to address potential risks associated with AI-powered systems. It also emphasizes the importance of controlling data quality, system validation, and continuous monitoring during operation.

Dekra plays a key role in promoting safe and reliable AI applications in road vehicles. The independent expert organization participates in regulatory discussions, possesses expertise in shaping best practices, and offers training, evaluation, and certification according to ISO 8800. This enables manufacturers to ensure that their AI systems meet the highest safety standards.

The higher the degree of automation, the more the causes of accidents shift from human error within the vehicle itself to human error in IT design behind the human-machine interface.

Dr. Thomas Wagner, Traffic psychologist at Dekra

Ethical Questions About AI Systems in Vehicles

The use of AI systems in vehicles also brings challenges, such as ethical issues and the need to ensure fundamental protection requirements. "The higher the degree of automation, the more the causes of accidents shift from human error within the vehicle itself to human error in IT design behind the human-machine interface," explains Dekra traffic psychologist Dr. Thomas Wagner. This is because those developing the software for vehicle AI must make a multitude of decisions, such as determining the behavior of fully automated vehicles in the event of an unavoidable accident and addressing the question of who should potentially be harmed.

"At its core, this involves programmed decision routines to distribute the risk of occurring damages among the various parties involved in an accident," the Dekra expert continues. Automakers and policymakers alike face these moral dilemmas. After all, reaching a consensus on the principles of such decisions is particularly important for the public, as fully automated vehicles would otherwise be socially less accepted and used.

Moral Dilemmas Experimentally Investigated

To quantify societal expectations regarding the ethical principles of conflict behavior in fully automated vehicles, a group of scientists led by Briton Edmond Awad developed the "Moral Machine" experiment a few years ago. Participants had to choose their preferred option in unavoidable accident scenarios with two possible outcomes:

• The fully automated vehicle evades or
• the vehicle stays on its course.

The "Moral Machine" collection ultimately included nearly 40 million decisions from millions of people in 233 countries or regions. A total of 130 countries were grouped into three broad clusters: the Western cluster (North America and many European countries), the Eastern cluster (e.g., Japan and Islamic countries), and the Southern cluster (countries in Central and South America).

An intensive societal discussion and as much international consensus as possible will be indispensable in the future.

Dr. Thomas Wagner, Traffic psychologist at Dekra

Between the three clusters, significant differences were observed in some preferences. For example, the preference to spare younger individuals over older ones and individuals of higher status over those of lower status was significantly less pronounced in countries of the Eastern cluster compared to the Western cluster. Countries in the Southern cluster showed a noticeably weaker preference for sparing humans over pets compared to the other two clusters. Only the preference for sparing pedestrians over vehicle occupants and the preference for sparing lawful traffic participants over those displaying unlawful behavior appeared to be equally present across all clusters.

"The results of this experiment show how multifaceted and complex the questions are that need to be decided during programming," says Dr. Wagner. "An intensive societal discussion and as much international consensus as possible will be indispensable in the future."

Secure and Compliant Authentication in Laboratories

Secure & Contactless User Authentication at GMP/GLP