EU AI Act Takes Shape: New Rules Aim to Curb Risks in Generative AI

The European Union has finalized key provisions of its AI Act, a landmark regulation targeting risks posed by generative artificial intelligence systems, according to a draft document reviewed by Reuters. The law, which could become enforceable by 2025, mandates transparency requirements for AI developers and bans certain “unacceptable risks” like biometric surveillance without consent.

According to the European Commission, the framework will require companies deploying AI systems to conduct risk assessments and maintain documentation of their algorithms’ training data. “This is the first time we’re establishing a comprehensive legal framework for AI at the EU level,” said Commission Vice President Věra Jourová. “Our goal is to ensure technology serves people, not the other way around.”

What Are the Main Provisions of the AI Act?

The legislation classifies AI systems into four risk categories, with strict rules applying to “high-risk” applications such as facial recognition and hiring tools. The EU’s regulatory body, the European Artificial Intelligence Board, will oversee compliance and impose fines of up to 6% of global revenue for violations.

Generative AI models like ChatGPT and MidJourney face specific requirements under the law. Developers must disclose when content is AI-generated and provide users with the ability to opt out of data collection for training purposes. “Transparency is non-negotiable,” said European Parliament rapporteur Sophie Perin. “Users deserve to know when they’re interacting with an algorithm.”

How Does This Compare to Other Global Regulations?

The EU’s approach contrasts with the U.S. model, which relies on sector-specific rules rather than a unified framework. China’s AI regulations, meanwhile, focus more on content control than algorithmic accountability. A 2023 study by the Brookings Institution found that 78% of EU citizens support stricter AI oversight, compared to 52% in the U.S.

The AI Act also introduces a “risk-based” classification system that differs from the UK’s proposed AI Standards Framework. While the UK emphasizes voluntary guidelines, the EU’s legislation includes mandatory assessments for systems used in education, healthcare, and criminal justice.

What Are the Challenges in Implementation?

Industry groups have raised concerns about the law’s potential impact on innovation. The European Tech Chamber warned that compliance costs could stifle startups, while the AI Ethics Alliance called for more flexibility in enforcement. “We need to balance accountability with the need for technological progress,” said spokesperson Markus Klein.

Technical challenges also remain. Researchers at the Max Planck Institute note that auditing complex AI systems requires new methodologies. “Current tools can’t fully trace how neural networks make decisions,” explained Dr. Lena Hofmann. “We’re developing explainability frameworks that could help regulators understand these systems.”

What’s Next for AI Regulation?

The AI Act now moves to the European Parliament for final approval, with negotiations expected to continue through 2024. If passed, it would set a precedent for global AI governance, influencing similar efforts in the U.S., Canada, and beyond.

As the technology evolves, experts emphasize the need for ongoing dialogue. “This is just the beginning,” said MIT Media Lab researcher Dr. Aisha Patel. “We must keep updating these rules as AI capabilities advance, ensuring they remain effective without hindering progress.”