On 9 December 2023, the Council of the EU and European Parliament reached an agreement on the Artificial Intelligence Act, or AI Act, the first comprehensive rules on AI worldwide. The AI Act aims to ensure the protection of democracy, fundamental rights, the rule of law, and environmental sustainability from certain risks posed by AI systems.
The AI Act will apply to:
- public and private actors from EU countries and non-EU countries when their system is placed on the Union market or affects users in the EU,
- providers and deployers of high-risk AI systems
- providers of general-purpose AI models (certain provisions)
Manufacturers will have to assess which risk category their AI system falls under and apply the correct conformity path. The Act presents a solid methodology to identify the risk category.
AI unacceptable risks banned in the EU
Most AI systems will pose low to no risk. However, some AI systems can present significant risks to the safety and health of users, with the opacity of many algorithms potentially representing a threat. Consequently, the new AI Act bans AI systems whose risk is classified as “unacceptable”. Systems posing an unacceptable risk are systems that could, for example, manipulate behaviors, doing social scorings (e.g., grouping people by behaviors, personal characteristics, socioeconomic background, etc.), and biometric identification (e.g., facial recognition). Products with such AI systems are banned from the EU market.
Conformity assessment of AI systems
Excluding unacceptable risks, AI systems with risks categorised as “high-risk”, “limited” and “minimal” can be placed on the EU market – once compliant with the applicable requirements. Products that include an AI system classified as “high-risk” will have to be subject to a third-party conformity assessment and mandatory fundamental rights impact assessment. Limited-risk systems will have to comply with transparency requirements, which include informing the users that they are interacting with AI technologies. Indeed, the agreement includes that general-purpose AI (GPAI) systems, and the GPAI models they are based on, will have to adhere to such transparency requirements. The rules for high-impact GPAI models with systemic risk are stricter. When meeting certain criteria, high-impact GPAI models with systemic risk will have to conduct model evaluations, assess and mitigate systemic risks, ensure cybersecurity, perform adversarial testing, report serious incidents, and inform on their energy efficiency.
Role of new technologies in the current revisions
During 2023, the Commission announced the revision or implementation of different legislation. The EU regulatory framework needs to update laws and provisions to include new technological development and sustainability aspects. In this context, the Machinery Regulation will replace the current Directive in 2027. Among the new rules, the Machinery Regulation will cover risks related to emerging digital technologies and introduce requirements for the safe use of Artificial Intelligence systems. Similarly, the General Product Safety Regulation will replace the current Directive (GPSD) already by the end of 2024. The GPSR foresees better health and safety protection from new technologies and AI risks. The product liability legislation is also being replaced by new rules in view of new technologies, considering that many products have digital features. Indeed, the new Directive will also include an extended definition of ‘product’ to include digital manufacturing files and software.
The agreed text will now have to be formally adopted by both the Parliament and Council to become EU law. Parliament’s Internal Market and Civil Liberties committees will vote on the agreement in a forthcoming meeting.
Stay tuned and do not miss any news!
Do you want to enter the EU market? Contact us today and we will guide you through the EU and UK regulatory requirements!
European Council (2023) Agreement between the Council and the European Parliament makes EU liability rules fit for the digital age and circular economy. Retrieved on 21.12.2023
European Commission (2023) Artificial Intelligence – Questions and Answers. Retrieved on 21.12.2023.