After marathon negotiations spanning nearly three days, lawmakers in the European Union (EU) have finally reached a political agreement on a risk-based framework for regulating artificial intelligence (AI). The proposed AI law, which was introduced in April 2021, has undergone months of challenging three-way talks to reach a consensus. This landmark development confirms that a pan-EU AI law is on its way.
EU lawmakers have agreed on a risk-based framework to regulate artificial intelligence, making it the first of its kind in the world.
The triumphant European Parliament, Council, and Commission representatives held a press conference in the early hours of Saturday morning, hailing the agreement as a hard-fought milestone and a historic achievement. EU President Ursula von der Leyen also took to Twitter to celebrate the political agreement, referring to it as a “global first”.
A Prohibition on Controversial AI Use Cases
Although the full details of the agreement will be confirmed in the upcoming weeks when a final text is released, a press release from the European Parliament provides an overview of the agreed-upon terms. Notably, the deal includes a total ban on the use of AI for certain controversial purposes:
- Biometric categorization systems based on sensitive characteristics such as political or religious beliefs, sexual orientation, or race.
- Untargeted scraping of facial images from the internet or CCTV footage to create facial recognition databases.
- Emotion recognition in workplaces and educational institutions.
- Social scoring based on social behavior or personal characteristics.
- AI systems designed to manipulate human behavior to infringe upon free will.
- AI used to exploit the vulnerabilities of individuals due to age, disability, or social and economic status.
Law enforcement will still be allowed to use remote biometric identification technology in public places, but strict safeguards and narrow exceptions will limit its use, including a requirement for prior judicial authorization. The scope of the technology will be confined to specific crime categories.
Regulation of High-Risk AI Systems and GPAIs
The agreed package also includes provisions for high-risk AI systems, defined as those posing significant potential harm to health, safety, fundamental rights, environment, democracy, and the rule of law. Mandatory fundamental rights impact assessments will be required for these systems, extending to the insurance and banking sectors. AI systems used to influence the outcome of elections and voter behavior will also fall under the high-risk category.
In addition to high-risk systems, the agreement establishes a “two-tier” system of guardrails for general-purpose AI systems, including foundational models. Transparency requirements will be imposed on low-tier AI systems, necessitating technical documentation and the publication of detailed summaries related to training content to ensure compliance with EU copyright law. High-impact foundational models will face more stringent obligations, such as model evaluations, mitigation of systemic risks, adversarial testing, reporting of serious incidents to the Commission, cybersecurity measures, and energy efficiency reporting.
Phased Implementation and Provisions for Startups
The agreed timeline for the AI Act’s entry into force allows for a phased approach. Six months will be granted for the implementation of rules related to prohibited use cases, followed by an additional 12 months for transparency and governance requirements. All other requirements will take effect after 24 months, meaning that the full impact of the EU’s AI Act may not be realized until 2026.
In order to support startups and SMEs in the development and training of AI systems, national authorities will be encouraged to establish regulatory sandboxes and real-world testing environments. Non-compliance with the regulations can result in fines ranging from €7.5 million or 1.5% of turnover to €35 million or 7% of global turnover, depending on the severity of the infringement and the size of the company.
The final text of the AI Act will still need to undergo voting in the parliament and the Council before adoption, but the political agreement reached represents a significant breakthrough in the EU’s journey to regulate AI. The European Commission has already begun preparations to implement the agreed-upon terms, including the establishment of an AI Office to coordinate with Member State oversight bodies.
The EU’s ambitious AI regulations are not only significant for the bloc’s digital market but also set a precedent on the global stage as the first comprehensive international regulation for artificial intelligence.