EU Agrees on World's First 'AI Regulation Law'... Risk Classification and Transparency Enhancement

Regulation Agreement After 37-Hour Marathon Meeting
Exceptions for AI Use in Judicial Investigations and National Administration

The European Union (EU) has finally agreed on the world's first artificial intelligence (AI) regulatory bill after lengthy discussions.

Multiple foreign media outlets, including Bloomberg News, reported that the EU Executive Commission (EC), the European Parliament, and representatives of the 27 EU member states reached an agreement on the bill called the 'AI Act' after more than 37 hours of prolonged meetings on the 8th (local time). Some countries such as France, Germany, and Italy argued that certain provisions should be relaxed as they could be disadvantageous to their domestic companies, which delayed the agreement, but a conclusion was ultimately reached.

[Image source=Reuters Yonhap News]

The bill includes provisions to classify AI risks and enhance transparency. It also imposes fines on companies that fail to comply with the regulations. Additionally, it prohibits obtaining biometric information from the internet and security footage to build facial recognition databases that classify people based on sensitive characteristics such as political or religious beliefs, sexual orientation, and race.

However, real-time facial recognition technology necessary for law enforcement authorities to search for victims of human trafficking, prevent terrorist threats, and track suspects of crimes such as murder and rape is permitted. Furthermore, while regulating large language models (LLMs) like ChatGPT (OpenAI) and Bard (Google), broad exceptions will be made for AI used for national security and law enforcement purposes.

Companies operating in the EU will be required to disclose data and undergo rigorous testing. Companies introducing high-risk technologies such as autonomous vehicles and medical devices will be subject to more detailed legal scrutiny. Violations of the regulations may result in fines of up to 35 million euros or 7% of global revenue.

Thierry Breton, EU Commissioner for the Internal Market, evaluated the bill as striking a balance between protecting the rights of individuals and companies. In a statement, he said, "It took a long time to find a balance between protecting citizens' fundamental rights and maximizing the potential of AI to support law enforcement," adding, "We do not want mass surveillance in Europe."

The final agreement document has not yet been released. The New York Times (NYT) in the United States anticipated that discussions on technical details would continue. The draft bill will undergo official approval by the European Parliament and member states. It is expected to take two years from approval to enforcement. The EU plans to introduce national and pan-European regulatory agencies for AI regulation.

The EU began discussions on AI regulation when the executive branch, the Commission, proposed the draft bill in April 2021. During this process, the emergence of new technologies led to revisions of the bill's content. It is known that the initial version did not mention general-purpose AI supporting ChatGPT.