EU negotiators reach a political deal on the AI Act
After trilogue talks that ran into the early morning of 9 December 2023, the European Parliament and the Council of the EU reached provisional political agreement on the Artificial Intelligence Act, setting risk tiers for AI systems and separate rules for general-purpose models.
Why it mattered Companies operating in the EU had their first concrete compliance target for general-purpose models, and governments drafting their own AI rules had a finished text to work from.
For most of 2023 governments reached for the laws they already had. In March Italy’s data protection authority ordered ChatGPT stopped for Italian users under data protection rules. In July the American Federal Trade Commission opened an investigation into OpenAI under consumer protection law. Neither instrument had been written with AI in mind.
The autumn brought instruments made for the subject and short on force. President Biden signed Executive Order 14110 on 30 October, binding on federal agencies and revocable by a later president. The United Kingdom gathered twenty-eight countries and the European Union at Bletchley Park on 1 November for a declaration with no binding force at all.
What was settled in Brussels in the small hours of 9 December was a law. After trilogue negotiations between the European Parliament and the Council of the EU that ran through the night, the two institutions announced provisional political agreement on the Artificial Intelligence Act.
The Act sorts systems into tiers by the risk they are judged to carry, with obligations rising as the risk does, and adds a separate set of rules for the general-purpose models released that year. December’s agreement was political rather than final. Formal adoption followed, and the obligations were written to take effect in phases over the years after it.
It was the first comprehensive AI law proposed by a major regulator, which made it the document other governments read while drafting their own.