The Council of the European Union (the Council) adopted a general approach on the draft Artificial Intelligence Act (AI Act) aimed at ensuring a common high standard for the safety of AI systems introduced to the EU market (on 6 December 2022). The AI Act is a proposal for a Regulation issued by the European Commission in April 2021 and forms a key aspect of the EU policy to nurture a safe and lawful AI environment across the EU.

Compared to the European Commission’s proposal, the Council’s approach includes the following key changes:

• the definition of “AI” is narrowed to systems developed through machine learning approaches and logic and knowledge based approaches. This is to easily distinguish AI from simpler software systems. The European Commission will be empowered to further specify or update what is covered by the AI Act, by adopting implementing acts;
• in relation to prohibited AI practices, the amended proposal extends the prohibition on using AI for social-scoring to private actors;
• the prohibition on the use of AI systems that take advantage of the vulnerabilities of specific groups is expanded to include vulnerabilities related to the social or economic situation of individuals;
• the provisions banning the use of real-time remote biometric identification systems in public spaces by law enforcement authorities have been clarified to exempt uses strictly necessary for law enforcement purposes in exceptional cases;
• the list of high-risk AI systems is expanded to include new use cases in critical digital infrastructure and life and health insurance, but also excludes some use cases (e.g. “deep fake” detection by law enforcement, crime analytics, verification of authenticity of travel documents);
• the requirements for high-risk AI systems are clarified to make them more technically feasible, for instance, in relation to the technical documentation to demonstrate compliance with the AI Act and the quality of data;
• provisions are added to clarify the responsibilities of various actors in AI value chains used for development and distribution of AI;
• some of the requirements for high-risk AI systems can now also apply to general purpose AI in specific situations, to be determined by an implementing act taking into account such factors as specific characteristics of the systems, its impact on the rights and freedoms of individuals, technical feasibility and market and technology developments.

In relation to the scope of the AI Act, the Council clarified that it will not apply (other than the obligations relating to transparency) to AI systems used for the exclusive purpose of research and development, or to use by individuals for non-professional purposes.

There is a strengthening of requirements relating to the transparency of high-risk AI systems, including a new obligation for public entities to register the systems in an EU-wide database, and specific requirements in relation to transparency with respect to emotion recognition systems.

To promote innovation, the Council of the EU amended the provisions on the proposed AI regulatory sandbox, among other things clarifying the conditions for unsupervised real world testing of innovative AI systems.

After the European Parliament has formed its position on the AI Act (expected in early 2023), EU legislators can proceed with the trilogue negotiations about the final text of the proposed Regulation.

The press release is available here and the adopted text here.