EU Parliament Approves World-First AI Act: Setting Global Standards for Artificial Intelligence
Strasbourg, France — In a historic vote held on March 21, 2025, the European Parliament formally adopted the final text of the Artificial Intelligence Act, commonly referred to as the AI Act framework. This decisive vote follows months of intense trilogue negotiations between the Parliament, the European Council, and the European Commission, coupled with detailed committee reviews, culminating in the creation of the world’s first comprehensive legal framework for Artificial Intelligence. The passage of this landmark legislation marks a pivotal moment in global AI governance, positioning the European Union at the forefront of regulating this rapidly evolving technology.
The Act establishes a groundbreaking risk-based approach to regulating AI systems. Under this framework, AI applications are categorized based on their potential to cause harm, with obligations scaling proportionally to the identified risk level. The highest level of scrutiny is applied to high-risk AI systems, defined as those used in critical sectors where the stakes are highest. These include applications in healthcare (like medical diagnostics), law enforcement (such as predictive policing or risk assessment tools), recruitment, critical infrastructure management, and systems used in democratic processes.
Providers and users of high-risk AI systems will face stringent obligations under the new law. These requirements encompass rigorous conformity assessments before the systems can be placed on the market, robust data governance practices ensuring the quality and integrity of data used for training, detailed technical documentation, transparency obligations towards users, human oversight mechanisms, and strong cybersecurity measures. The aim is to ensure that AI systems deployed in sensitive areas are safe, reliable, transparent, and accountable.
Conversely, the Act sets significantly lighter rules for lower-risk applications. AI systems deemed to pose minimal or limited risk, such as recommender systems or spam filters, will face fewer regulatory hurdles. For systems interacting directly with humans, like chatbots, transparency requirements will necessitate informing users that they are interacting with an AI. This tiered approach is designed to foster innovation while mitigating the most significant potential harms associated with AI deployment.
Key Provisions and Prohibited Practices
A central element of the AI Act is the prohibition of certain AI practices deemed to pose an unacceptable risk to fundamental rights and democratic values. These include specific bans on certain intrusive AI uses. Prominently featured among these prohibitions is the ban on the use of real-time remote biometric identification systems, such as real-time facial recognition, in public spaces. This ban is intended to protect citizens’ privacy and freedom of assembly, although the Act does provide for limited exceptions under strictly defined circumstances, such as the search for specific victims of crime or the prevention of a terrorist threat, subject to judicial authorization and strict necessity and proportionality requirements.
Other prohibited practices outlined in the Act include AI systems used for social scoring by governments or for manipulative techniques that can distort behaviour in a manner that causes significant harm. The legislation also bans AI used to exploit the vulnerabilities of specific groups due to their age, disability, or social or economic situation.
Specifics for Generative AI
Recognizing the emergence and rapid development of generative AI models, such as those powering large language models (LLMs) and image generators, the Act includes specific provisions tailored to these systems. Providers of generative AI models will face mandates for transparency and data governance. This includes the requirement to clearly label content generated by AI as artificial, providing summaries of the copyrighted data used for training purposes, and implementing safeguards against generating illegal content. These rules are aimed at addressing concerns around deepfakes, misinformation, and intellectual property rights in the context of powerful generative AI.
The Journey to Approval and Global Impact
The approval of the AI Act was the culmination of a complex legislative process that began with the European Commission’s proposal in April 2021. The journey involved extensive debate and amendments within the EU Parliament and the Council, leading to the intensive trilogue negotiations that reconciled the differing positions. These multi-institutional discussions and subsequent committee reviews were crucial in shaping the final text approved in Strasbourg on March 21, 2025.
The EU’s AI Act is widely seen as setting a global precedent for AI regulation. Its risk-based framework and specific provisions are expected to influence regulatory approaches in other jurisdictions around the world, potentially creating a ‘Brussels Effect’ where global companies adhere to EU standards to operate in the large European market. While the Act is now formally approved, it will enter into force gradually, with different provisions becoming applicable over the coming months and years, allowing time for businesses and authorities to adapt to the new rules. This landmark legislation represents a significant step towards ensuring that AI development and deployment in the EU serve humanity, adhering to fundamental rights, safety, and ethical principles.
