The European Commission proposes the AI Act
A risk-tiered draft banning practices such as social scoring outright; three years of negotiation followed before it became binding law.
- Government & policy
- Major
The European Commission published COM(2021) 206, a draft regulation laying down harmonised rules on artificial intelligence across the European Union — the first proposal anywhere for comprehensive, horizontal AI law rather than sector-specific rules or voluntary guidance.
The draft sorted AI systems into risk tiers rather than regulating the technology as a whole. A short list of “unacceptable risk” practices was to be banned outright, including AI that uses subliminal techniques to manipulate behaviour, that exploits the vulnerabilities of children or people with disabilities, that enables government social scoring of citizens, and — with narrow law-enforcement exceptions — real-time remote biometric identification in public spaces. A larger category of “high-risk” systems, covering uses in areas such as employment, credit scoring, law enforcement and critical infrastructure, would face mandatory requirements on data governance, documentation, human oversight and conformity assessment before they could be placed on the market. Systems judged low or minimal risk would face only transparency obligations, such as disclosing that a user is interacting with a chatbot.
The proposal predated the current wave of large generative models by more than a year and a half, and its drafting reflected that: it was built around discrete, task-specific AI systems rather than general-purpose foundation models, a gap negotiators had to address later as ChatGPT and its successors made the omission conspicuous. The Commission’s proposal was the opening move in a legislative process that ran for three more years, through committee amendments, a political trilogue and last-minute additions covering general-purpose AI, before the European Parliament passed a final text in March 2024 and the Council of the EU gave its formal sign-off two months later. It remains the most-cited model internationally for what comprehensive AI regulation might look like, whether invoked approvingly or as a cautionary example of regulatory complexity.