Reed Smith In-depth

Key takeaways:

  • The aim of the Act is to regulate the development and use of AI systems by providing a horizontal framework of obligations for AI developers and AI users ranging from mere transparency requirements to obtaining a comprehensive CE marking. This framework is based on a risk-categorisation of AI systems (low, limited, high) including a prohibition on certain types of AI that pose unacceptable risk.
  • The final version of the Act also addresses “general-purpose AI “(GPAI), or models which can be used for a variety of tasks, with the most powerful GPAI models - referred to as 'systemic' – subject to additional requirements.
  • Special attention needs to be drawn to the requirement imposed by the Act to GPAI to ‘comply with EU copyright law’ which applies on a territorial and on an extraterritorial basis to “ensure a level playing field”. This is unprecedented with respect to a right that is inherently territorial and a significant step towards trying to establish EU copyright law as a global standard. 
Artificial intelligence concept - AI brain floating above hands

What is the EU AI Act?

In April 2021, the European Commission tabled a proposal for the AI Act (the “Act”). In typical Brussels fashion, the Act was then under discussion for over two years until December 2023 when a political agreement was finally reached. The Act was approved by the EU Parliament in March and the same version was formally approved by the Council during the plenary session on May 21st, with publication still pending in the Official Journal of the Union.

The aim of the Act is to regulate the development and use of AI systems by providing a horizontal framework of obligations for AI developers and AI users ranging from mere transparency requirements to obtaining a comprehensive CE marking. This framework is based on a risk-categorisation of AI systems (low, limited, high) including a prohibition on certain types of AI that pose unacceptable risk.  The final version of the Act also addresses general-purpose AI (GPAI), a term that has been preferred to ‘foundation models’ to describe models which can be used for a variety of tasks, with the most powerful models - referred to as 'systemic' - subjected to additional requirements.

Crucially, the proposed Act is a regulation, meaning that it will be directly applicable in the European Union (EU) Member States and will not need to be transposed into national law. With its entry into force, the regulation will then become part of national law and will be enforceable through national courts of each Member State.

When will the Act come into force?

The Act will come into effect two years after its publication (in 2026), with some particular provisions taking effect sooner: prohibitions will apply after six months, while most rules regarding general-purpose AI, governance, notified bodies, and sanctions will apply after twelve months. For product components covered by sectoral legislation, such as toys, cars or medical devices, and for GPAI model already on the market at the date of the Act, the implementation deadline will be thirty-six months. See our AI timeline on reedsmith.com.