The European Parliament and Council have struck a historic deal on the Artificial Intelligence Act, prioritizing safety, EU values, and innovation. The legislation outlines regulations for high-risk AI, prohibits certain practices, and introduces a governance framework. Emphasizing transparency and fundamental rights, the EU's AI Act has the potential to become a global standard for AI regulation.
After three days of negotiations, the European Parliament and the Council presidency have come to a provisional agreement on the artificial intelligence act. The innovative approach prioritizes the safety of AI systems, EU values, and fundamental rights. Its objectives are to guarantee responsible AI use, encourage investment, and promote creativity throughout the continent.
About The Act
The regulation strives to prevent high-risk AI from undermining fundamental rights, democracy, the rule of law, and environmental sustainability while simultaneously fostering innovation and positioning Europe at the forefront of the industry.1
The compromise agreement, aligned with OECD standards, ensures a precise definition of AI systems and establishes a regulatory framework applicable to areas covered by EU law. While it covers various areas, there are specific exemptions for national security authorities, particularly for defense or military applications and non-professional or research purposes.
Requirements For Risky Systems
AI systems that present a high risk to human health, safety, the environment, democracy, and the rule of law will be subject to special regulations. Members of the European Parliament have ensured the inclusion of the banking and insurance industries in the mandatory fundamental rights impact assessment. Voter behavior and election-related AI systems are classified as high-risk. In cases where decisions made by high-risk AI systems impact citizens' rights, individuals have the right to file complaints and request explanations.
Prohibited AI Practices
In light of potential risks to democracy and citizens' rights, co-legislators have prohibited certain AI applications. These include social rating based on personal traits, emotion awareness in the workplace and education, the use of sensitive attributes for biometric classification, employing faces for unintended face recognition database scraping, leveraging AI systems to influence human behavior against free will, and exploiting vulnerabilities based on age, disability, or social and economic standing.
Exceptions For Law Enforcement
The Commission has revised its AI proposal for law enforcement to underscore the benefits of AI while prioritizing data privacy top. In emergency situations, high-risk AI tools can be deployed without undergoing initial assessments for conformity as long as a mechanism designed to safeguard fundamental rights is activated. The agreement sets strict guidelines for law enforcement's exceptional use of real-time remote biometric identification. Exceptions are limited to cases such as preventing genuine, present, or foreseeable threats (e.g., terrorist attacks), assisting victims of specific crimes, and pursuing those responsible for the most severe crimes.
The Purpose
In addition to outlining the integration of General-Purpose AI (GPAI) into high-risk systems, the draft agreement establishes guidelines for foundation models. These large systems exhibit capabilities in content creation and coding. Notably, the agreement mandates transparency compliance before market release. Due to potential systemic risks across the value chain, "high impact" foundation models, characterized by advanced capabilities and extensive data training, are subject to a stricter framework.
A Fresh Framework for Governance
With the endorsement of a scientific panel, new regulations establish an AI Office tasked with supervising GPAI model rules. The office is responsible for monitoring safety concerns, identifying high-impact models, evaluating GPAI capabilities, and offering guidance to member states. The AI Board, which continues to be staffed by representatives of member states, plays a crucial role in providing guidance and facilitating the implementation of regulations. In an advisory forum, technical expertise is contributed by academia, civil society, small and medium enterprises, industry, and start-ups.
Sanctions and Penalties
Sanctions under the AI Act are determined based on global annual turnover in the previous financial year or a predefined amount, whichever is higher. The fines are structured as follows:
- €35 million or 7% for banned AI applications,
- €15 million or 3% for the violation of AI Act obligations, and
- €7.5 million or 1.5% for inaccurate information.
In the event of violations, the provisional agreement imposes caps for SMEs and start-ups. It underscores the provision for individuals or groups to file complaints with the market surveillance authority, ensuring proper handling in accordance with established procedures.
Protecting Fundamental Rights and Ensuring Transparency
Before high-risk AI systems are released onto the market, the provisional agreement mandates an evaluation of the impact on fundamental rights. Introducing a registration requirement in the EU database for artificial intelligence systems that pose a high risk, this measure extends to certain public entities as well. The new provisions also emphasize the obligation for users of emotion recognition systems to alert individuals when they are being exposed to such technology, further enhancing transparency in their use.
Policies That Foster Innovation
The AI legal framework outlined in the provisional agreement is crafted to be evidence-based and to foster innovation. Initially designed for controlled testing, AI regulatory sandboxes are now open to real-world testing, enabling the testing of AI systems in practical scenarios, subject to additional safeguards. Specific exclusions and measures have been incorporated to mitigate administrative burdens on smaller companies.
What's Next?
The final draft of the provisional agreement will be crafted through technical-level studies. Subsequently, the agreement will be submitted to representatives of member states for endorsement. Before formal legislative approval, the entire text will undergo comprehensive legal and linguistic verification and revision.
The AI Act is scheduled to be applied two years after its entry into effect, with specific provisions having exceptions.2
Conclusion
The EU's AI Act aims to promote safe and trustworthy AI development within the EU's single market for both public and private entities, employing a 'risk-based' approach with stricter rules for higher-risk AI applications. It establishes a consistent, horizontal legal framework for AI with the goal of ensuring legal certainty.
As the world's inaugural legislative proposal of its kind, it has the potential to set a global AI regulation standard, solidifying the European tech regulation approach internationally.
Footnotes
1. Artificial Intelligence Act: deal on comprehensive rules for trustworthy AI, 2023
2. Artificial intelligence act: Council and Parliament strike a deal on the first rules for AI in the world, 2023
References
A European approach to artificial intelligence. (n.d.). Retrieved from European Commission: https://digital-strategy.ec.europa.eu/en/policies/european-approach-artificial-intelligence
Artificial intelligence act: Council and Parliament strike a deal on the first rules for AI in the world. (2023, December 9). Retrieved from European Council - Council of the European Union: https://www.consilium.europa.eu/en/press/press-releases/2023/12/09/artificial-intelligence-act-council-and-parliament-strike-a-deal-on-the-first-worldwide-rules-for-ai/
Artificial Intelligence Act: deal on comprehensive rules for trustworthy AI. (2023, December 9). Retrieved from European Parliament: https://www.europarl.europa.eu/news/en/press-room/20231206IPR15699/artificial-intelligence-act-deal-on-comprehensive-rules-for-trustworthy-ai
Cover Note by Council of the European Union No. 8115/21. (2021, April 23). Retrieved from European Council - Council of the European Union: https://data.consilium.europa.eu/doc/document/ST-8115-2021-INIT/en/pdf
Note by the Council of the European Union No. 14954/22. (2022, November 25). Retrieved from European Council - Council of the European Union: https://data.consilium.europa.eu/doc/document/ST-14954-2022-INIT/en/pdf
Your life online: How is the EU making it easier and safer for you? (n.d.). Retrieved from European Council - Council of the European Union: https://www.consilium.europa.eu/en/your-online-life-and-the-eu/#group-section-trustworthy-AI-uoApCQWI8w
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.
