Artificial intelligence (AI) is changing the game for motor vehicles through the development of self-driving systems that autonomously control the driving functions of vehicles they are installed in. While autonomous/automated vehicles (AVs) may be appealing, especially to those who make long journeys, the use of AI in autonomous vehicles also introduces novel health and safety risks, such as AVs going off the road or crashing into objects or other vehicles and causing accidents.
Fueled by these risks, there are increasing proposals for laws to set out safety requirements for autonomous vehicles, particularly in the US. Across the pond, the EU AI Act, which governs the development, deployment, and use of AI systems within the European Union and classifies them according to their risk levels, will also have implications for autonomous vehicles. In this blog post, we outline how the AI Act addresses autonomous vehicles, navigating the existing legislative landscape.
Key Takeaways
AVs have a sophisticated suite of sensors, cameras, and advanced computing systems that empower them to navigate roads with little to no human input. These vehicles represent a new era in transportation, capable of safely maneuvering through an ever-changing environment with precision and awareness.
The degree of autonomy within these vehicles is classified according to their capabilities and the extent to which they require driver engagement. This spectrum ranges from partial automation, where the vehicle assists with certain tasks, to full automation, where the vehicle assumes all driving functions.
Through AI, AVs gain the ability to interpret sensory information, enabling them to perceive their surroundings. This includes recognizing obstacles, interpreting traffic signals, and navigating complex environments. AI algorithms are responsible for decision-making that allow AVs to execute maneuvers, route planning, and real-time adjustments based on dynamic road conditions. Moreover, AI systems can learn from vast amounts of data derived from driving scenarios, helping to improve the vehicles’ driving strategies.
The use of AI in AVs is not limited to essential driving functions, however – it also extends to enhancing the passenger experience through various non-essential, entertainment-focused applications. For instance, AI-driven interfaces can personalize the in-cabin experience, adjusting environmental settings like lighting and temperature based on passenger preferences. Additionally, AI enables sophisticated infotainment systems that can provide real-time news updates, recommend music or podcasts based on listening history, or even suggest points of interest along a route.
There are two primary regulations applicable to AVs in the EU: (1) the Type-Approval Framework Regulation (TAFR) and (2) the General Safety Regulation (GSR). According to these regulations, AVs, and their components, like many other products in the EU market, must priorly undergo a type-approval process to ensure that the type of vehicle, system, component, or separate technical unit satisfies the relevant administrative provisions and technical requirements.
The TAFR and GSR put forward a comprehensive set of safety and environmental mandates, including crashworthiness standards, electronic stability control protocols, and fuel efficiency benchmarks. These requirements ensure that all motor vehicle types, particularly AVs, uphold the highest safety and environmental stewardship standards. Manufacturers and other stakeholders in the automotive industry must align their AV models and components with the outlined requirements in these regulations to enter the EU market.
The EU AI Act takes a risk-based approach to regulation, with high-risk AI systems (HRAIS) having the most stringent obligations. High-risk systems are those with the potential to significantly affect people’s health, safety, or fundamental rights, and AI systems deployed on or in connection with AVs are a great example of this: if an AI system fails in an AV, it could lead to real-world consequences like traffic accidents or breaches of privacy.
While many AI systems used in AVs will likely fall under the HRAIS category, the EU AI Act recognizes that AVs are already regulated by other Union legislation, such as TAFR and GSR. As such, the Act aims to ensure that the governance of these systems is aligned with the existing rules on the regulation of vehicles by not directly introducing rules or requirements for these and only creating a bridge between the sectoral requirements and the HRAIS requirements at a high-level. Therefore, although these systems will be classified as high-risk, they will primarily be governed by their sectoral legislation and the standard requirement or obligation regime for HRAIS will not apply to these systems.
For the remaining systems used in AVs, that are not covered by sectoral regulation, the standard approach to determining risk level will be applied, meaning that it must be explored whether the system falls under the high-risk use cases (Annex III) or operates in ways that are subjected to transparency requirements (Article 52). The categorization of AI systems deployed on or in connection with AVs under the EU AI Act can be summarized as follows:
There are to cumulative conditions that the EU AI Act uses to classify AI systems embedded in products governed by specific Union harmonization legislation as HRAISs:
Both the TAFR and the GSR are part of the list mentioned above of Union harmonization legislation in Annex II of the EU AI Act, and given the role of AI systems in AVs as safety components, addressing both the safety of vehicle occupants and external entities like pedestrians, they would be classified as hgh-risk.
In addition, the TAFR requires vehicles to undergo a conformity assessment by bodies known as ‘technical services’, meaning many AV-related AI systems will be classified as HRAIs under the EU AI Act due to the reference to the Union harmonization legislation.
Article 2 of the EU AI Act outlines the provisions for HRAIS governed by specific Union harmonization legislation, including AVs. Accordingly, Article 84 (evaluation and review by the Commission) Article 53 (AI regulatory sandboxes) will apply to HRAIS of AVs.
There will also be specific amendments for the TAFR and the GSR under Article 80 and Article 82, which instruct the Commission to consider HRAI requirements in future adoptions of legislative pieces. In other words, the EU AI Act’s provisions regarding HRAIS, including the requirements and obligations, will not apply to AVs. Hence, this special regime can be outlined by 3 pillars:
For the AV-related AI systems that are not covered by the TAFR and GSR, the usual approach to risk classifications under the EU AI Act may apply if they meet the conditions outlined in Annex III. Additionally, if these AI systems perform the tasks mentioned in Article 52, they could be subject to specific transparency obligations. Alternatively, systems that present minimal risk may be regulated through voluntary codes of conduct according to Article 69.
The exact obligations of stakeholders in the automotive sector may be clarified with future changes to sectoral legislation on AVs. Therefore, it is not yet clear whether these obligations will fully comply with the obligations of stakeholders addressed in the EU AI Act.
Consequently, certain actors in the supply chain may be affected in different ways:
The EU AI Act represents a significant shift for automotive sector stakeholders, pointing towards harmonization of AV directives with HRAI requirements, which require the establishment of risk management and quality management systems. Holistic AI’s expertise in AI risk assessment could offer crucial clarity and strategic foresight into the ramifications of the EU AI Act on AVs and the assimilation of AI technologies within this sphere.
Schedule a call to find out more about how Holistic AI can help you with EU AI Act readiness.
Written by Osman Gazi Güçlütürk, Legal and Regulatory Lead at Holistic AI; Bahadır Vural, Legal Researcher at Holistic AI.
Last updated 20 March 2024.
Govern AI to continue scaling and adoption
DISCLAIMER: This blog article is for informational purposes only. This blog article is not intended to, and does not, provide legal advice or a legal opinion. It is not a do-it-yourself guide to resolving legal issues or handling litigation. This blog article is not a substitute for experienced legal counsel and does not provide legal advice regarding any situation or employer.
Schedule a call with one of our experts