The European Commission, the European Parliament and negotiators from 27 member states reached an agreement on Friday (December 8) on the world’s first artificial intelligence (AI) regulatory rules, including how to limit the government’s use of AI in biometric tools. , and how to manage generative AI systems such as ChatGPT.
“A historic moment!” After more than 33 hours of marathon negotiations, Thierry Breton, the EU’s executive commissioner responsible for the EU’s internal market and industrial policy, announced before midnight on Friday that the EU has become the first to formulate regulations for the use of AI. A continent with clear rules.
Breton said at
The draft legislation still needs to be formally approved by EU member states and parliaments. But this agreement marks a key step towards AI policy in the Western world. Currently, the U.S. Congress has not taken meaningful action on AI technology, so EU legislation will set the tone for the regulation of this rapidly developing technology.
The EU aims to enact strict regulatory measures on AI technology for the first time outside Asia.
Under the agreement, basic models such as ChatGPT and the General Artificial Intelligence System (GPAI) must comply with transparency obligations before being marketed, including the preparation of technical documentation, compliance with EU copyright law and the dissemination of detailed summaries of relevant training content.
High-impact basic models with systemic risks must conduct model evaluations, assess and mitigate systemic risks, conduct adversarial testing, report serious events to the European Commission, ensure network security, and report on their energy efficiency, etc.
The agreement limits the government to only conducting real-time biometric surveillance in public places, such as for victims of specific crimes, to prevent foreseeable threats such as terrorist attacks, and to find suspects involved in the most serious crimes. The agreement also bans the use of AI to manipulate cognitive behavior and infer people’s political or religious beliefs, sexual orientation or race based on sensitive biometric characteristics.
Consumers will have the right to lodge a complaint and receive a meaningful explanation, while fines for non-compliance will range from €7.5 million (approximately $8.1 million) or 1.5% of turnover to €35 million (approximately $37.8 million) or 7% of global turnover. % varies.
“We don’t want any mass surveillance in Europe,” Breton said in a statement on Saturday.
Civil society is divided. Cecilia Bonefeld-Dahl, director general of the business and technology group DigitalEurope, criticized these rules as another burden on businesses, “We have reached an agreement, but at what cost?”
Daniel Friedlaender, head of the European office of the Computer and Communications Industry Association (CCIA), which lobbies on behalf of the tech industry, said: “Today’s political agreement marks the start of critical work on the Artificial Intelligence Act, but Details are still lacking.”
European Digital Rights, which advocates privacy protection, criticized the organization. Ella Jakubowska, the organization’s senior policy adviser, said that this is the first time the EU has taken measures to legalize real-time public facial recognition. , “Despite parliamentary efforts to limit the damage, the accompanying package on biometric surveillance and profiling has been lukewarm at best.”
Once formally approved, the legislation is expected to come into force as soon as early next year and should be implemented two years later.
(This article refers to reports from Reuters and the Associated Press)