The consensus between EU member states and the European Parliament signifies Europe’s ambition to lead as a global standard-setter in the digital realm. Thierry Breton, European Commissioner, lauded the day as historic, indicating Europe’s realization of the significance of its role on the world stage.
Under the agreement, foundational models like ChatGPT and general-purpose AI systems (GPAI) must adhere to strict transparency requirements before entering the market. This will include crafting technical documentation, conforming to the EU’s copyrights laws, and providing detailed summaries of datasets used for training.
High-risk foundational models will undergo rigorous scrutiny, involving model evaluations, systemic risk assessments and mitigation, adversarial testing, incident reporting to the European Commission, cybersecurity measures, and energy efficiency reporting. GPAI systems that carry systemic risks can align with new regulations through codes of practice.
The use of real-time biometric surveillance within public spaces by authorities is to be restricted, permitted only in cases involving specific crime victims, the prevention of imminent, immediate, or predicted threats such as terrorist acts, and in searches for individuals suspected of serious criminal offenses.
The agreement also outlaws practices perceived as manipulative or invasive, such as cognitive and behavior manipulation, harvesting facial images from the internet or CCTV footage, social scoring systems, and biometric categorizations to ascertain political, religious, philosophical beliefs, sexual orientation, or race. Additionally, consumers will have the right to lodge complaints and receive meaningful explanations to grievances. Violations carry hefty penalties ranging from 7.5 million euros or 1.5% of turnover to 30 million euros or 7% of global turnover.
These ambitious regulatory efforts by the European Union arrive as companies like OpenAI, which counts Microsoft as an investor, continue to explore innovative applications of AI, sparking both admiration and concern. Alphabet, the parent company of Google, has introduced a new AI model, Gemini, challenging OpenAI’s domain. The EU’s AI legislation may now set a precedent for other governments, offering an alternative to the light-touch approaches seen in the U.S. and stricter regulations encountered in China.
