Artificial Intelligence Act
European Parliament adopts its first-reading position on the AI Act
View full image ↗ How was this image made?
The creative brief sent to the image model. These are instructions, not a record of what actually happened.
Original prompt & settings (JSON) ↗The vote was 523 to 46 with 49 abstentions — and the AI Act still was not law when Parliament adjourned that evening.
What Strasbourg actually voted on
Meeting in Strasbourg on 13 March 2024, the European Parliament approved its first-reading position on the proposed Artificial Intelligence Act by 523 votes to 46, with 49 abstentions.
The text MEPs endorsed reflected the provisional political agreement Parliament and Council negotiators had reached in December 2023, after committee approval in February and an undertaking from a Council representative to accept Parliament's position without amendments.
Co-rapporteurs Brando Benifei and Dragoș Tudorache had led Parliament's work on the file since the Commission proposed the regulation in April 2021.
Why 'passed by the European Union' compresses the story
The listing describes the Act as passed by the EU on this date. That day only one of the two co-legislators acted: Parliament fixed its position and cleared its part of the ordinary legislative procedure.
Formal Council approval followed on 21 May, unanimously, with 27 member states in favour and none against or abstaining. Publication in the Official Journal came on 12 July, and the regulation entered into force twenty days later, on 1 August.
Entry into force and application are separate legal dates, and the Act's obligations were phased in on different schedules.
The summaries that oversimplify the text
The familiar four-tier risk diagram is a summary, not a complete description: the regulation handles prohibited practices, high-risk systems, transparency duties and general-purpose models through different mechanisms.
Real-time remote biometric identification in public spaces for law enforcement is not absolutely banned. Article 5 permits narrowly defined uses subject to necessity, safeguards and prior judicial or independent administrative authorisation.
A training-compute threshold of 10^25 floating-point operations creates a rebuttable presumption that a general-purpose model has high-impact capabilities; the Commission may also designate models on other criteria.
Sources
Researched 24 Aug 2026 5 sources not yet audited
How this was checked
Researched from the web into a fact sheet and rewritten from that sheet. This article has not yet had an independent model audit. How the pipeline works →
What the sources leave uncertain
- The March 13 vote is fully documented, but calling it the date the Act was “passed by the European Union” compresses a multi-stage legislative process.
- Descriptions such as “the world's first comprehensive AI law” are institutional characterizations. The Council called it the first law of its kind worldwide, but jurisdictions had already enacted narrower AI rules.