The European Parliament (EP) and the Council reached on 8 December a political deal on the final version of the Artificial Intelligence Act (AIA).
The future AI rulebook will keep the risk-based approach proposed by the European Commission, while introducing significant changes to the list of banned AI practices, the regulation of foundation models, the envisaged EU governance structure, as well as the measures in support of innovation (such as sandboxes).
Now the EU co-legislators will need to approve the final text (subject to additional work at technical level) before its formal adoption.
This Cullen International Cheat Sheet provides a summary of the highlights in the provisional agreement as described in the press releases published by the EP and the Council. The summary reflects the changes compared to the initial Commission proposal.
(Free access for all)
Clients of our European Digital Economy service can access it directly here.
Related Infographics:
April/May 2025:
- This timeline outlines the milestones in the implementation of the Artificial Intelligence Act (AI Act) until August 2027:
Download our EU AI Act Timeline
- This cheat sheet provides a summary of the core provisions on General-purpose AI (GPAI) models under the AI Act
Download our AI Act Cheat Sheet on GPAI Models
more news
11 September 26
Have EU countries implemented the e-Evidence Package?
Cullen International’s latest Benchmark tracks implementation of the e-Evidence Regulation and transposition of the accompanying Directive on legal representatives across 26 EU member states.
10 September 26
Digital Networks Act recitals now directly linked to articles in the Cullen Legislation Navigator
Digital Networks Act (DNA) recitals now link to articles in Cullen international's Legislation Navigator online tool.
09 September 26
Insights into the proposed EU framework for mobile satellite services (MSS) in the 2 GHz band
Cullen International’s new Tracker makes a comparison between the European Commission’s 2 GHz band MMS proposals and the existing framework.