The European Union's Artificial Intelligence Act, the world's first comprehensive legal framework for artificial intelligence, received its final approval from the Council of the European Union on May 21, 2024. This landmark legislation aims to foster the development and adoption of safe and trustworthy AI systems across the EU's single market while upholding fundamental rights. The approval concludes a legislative process that began with the European Commission's proposal in April 2021, followed by a provisional agreement between the Council and the European Parliament in December 2023, and subsequent endorsement by the Parliament on March 13, 2024.

The AI Act is designed with a risk-based approach, categorizing AI systems based on their potential to cause harm. It places strict obligations on high-risk AI applications, while prohibiting certain systems deemed to pose an unacceptable threat to human safety and rights. This regulatory precedent is expected to influence AI governance frameworks globally, setting a benchmark for ethical and responsible AI development. Businesses and organizations operating or offering AI systems within the EU will need to comply with the new rules, regardless of their origin.

Key aspects and implications of the EU AI Act include:

  • Prohibited AI Systems: The Act bans AI systems that manipulate human behavior, exploit vulnerabilities, or are used for social scoring by governments, among others. These prohibitions will apply six months after the Act's entry into force.
  • High-Risk AI: Systems used in critical infrastructure, education, employment, law enforcement, migration management, and democratic processes are classified as high-risk. These systems will face stringent requirements regarding data quality, human oversight, transparency, robustness, and conformity assessments. Obligations for high-risk AI will largely apply 36 months after the Act's entry into force.
  • General-Purpose AI (GPAI): The Act introduces specific rules for GPAI models, including those with significant systemic risk, such as large language models. Developers of GPAI models must comply with transparency requirements, technical documentation, and adhere to EU copyright law. These provisions will become applicable 12 months after the Act's entry into force.
  • Sandboxes and Innovation: To support innovation, the Act provides for regulatory sandboxes and real-world testing, allowing SMEs and startups to develop and train innovative AI before full market deployment.
  • Enforcement and Penalties: Non-compliance can lead to significant fines. For instance, violations related to prohibited AI practices could result in penalties up to €35 million or 7% of a company's global annual turnover, whichever is higher.

Following its publication in the Official Journal of the European Union, the AI Act will enter into force 20 days later. Its provisions will then apply gradually, with some rules coming into effect within six months and others taking up to 36 months. Member states now face the task of establishing national frameworks, supervisory authorities, and guiding businesses through the transition, emphasizing the need for robust technical standards and implementation guidelines to ensure effective and harmonized application across the Union.