European Parliament Approves Strict AI Act With New Foundation Model Rules
The European Parliament approves its draft version of the EU Artificial Intelligence Act, paving the way for final negotiations. The legislation targets generative AI and expands bans on high-risk surveillance tools.
The European Parliament approves its draft version of the EU Artificial Intelligence Act, marking a major step toward comprehensive AI regulation. European institutions now enter negotiations to finalize the text, meaning the law applies at the earliest in 2025. If adopted, the rules apply to any provider or deployer of in-scope AI systems that operate within or produce effects in the EU, regardless of where the company is established.
The Parliament's text focuses heavily on promoting trustworthy AI by expanding existing restrictions. Lawmakers vote to ban AI systems that use remote biometric identification in public spaces, emotion recognition technology, and predictive policing systems. Furthermore, the definition of high-risk AI expands to include systems designed to influence voters and the recommender algorithms used by very large online platforms.
A significant addition in this version involves new requirements for foundation models, which are broad AI systems trained on massive datasets that power generative AI tools. Providers of these foundation models face specific obligations regarding their development and deployment. This updated approach builds on the European Commission's original 2021 proposal, which initially categorized AI systems into unacceptable, high, and limited risk tiers.