EU lawmakers back tighter AI rules for foundation models

A committee in the European Parliament has approved updated amendments to the EU’s AI Act that would add new transparency and risk-assessment duties for developers of foundation models. The move is part of the bloc’s broader effort to bring more oversight to powerful artificial intelligence systems used across consumer and business products.

The proposal would require providers of these models to disclose more information about how their systems are built and tested, while also documenting potential risks. Lawmakers say the goal is to improve accountability and reduce the chance of harm from advanced AI tools that can generate text, images, code and other outputs at scale.

The committee vote is an important step, but it does not finalize the law. The amendments still need further negotiations and approvals before becoming binding EU policy. The outcome will shape compliance expectations for major AI developers operating in Europe and could influence regulatory debates far beyond the bloc.

The latest changes reflect rising concern in Brussels over how quickly foundation models are spreading and how little is known about their training data, safety controls and downstream impacts. Supporters argue the rules are necessary to ensure innovation does not outrun public safeguards.