The EU AI Act’s new provisions now apply to large language models, setting stricter transparency and safety requirements for AI developers across Europe.

The European Union has moved a critical step forward by making the AI Act’s new provisions on large language models legally enforceable, meaning AI developers must now meet tighter transparency and safety standards across the bloc.

What the AI Act Now Covers

The latest amendment expands the scope of the AI Act to include large language models (LLMs) such as chatbots and generative text tools. Providers must conduct risk assessments, document training data sources, and implement robust monitoring to prevent harmful outputs.

Key Compliance Requirements

  • Publish a model card detailing capabilities, limitations, and intended use cases.
  • Maintain a log of data provenance and any third‑party datasets used for training.
  • Implement real‑time monitoring to detect disallowed content and bias.
  • Allow regulators to request audits and access to source code under controlled conditions.

Non‑compliant firms face fines of up to 6% of global annual turnover, mirroring penalties for other EU data protection breaches.

Impact on AI Companies

Both established tech giants and emerging startups will need to allocate resources for documentation, testing, and ongoing compliance checks. Some smaller firms have expressed concern about the administrative burden, while larger players are already adapting their development pipelines to meet the new rules.

Enforcement Timeline

National supervisory authorities will begin inspections in early 2027, with a phased rollout that prioritizes high‑risk applications such as finance, healthcare, and public administration.

"The enforceability of these provisions signals the EU’s commitment to trustworthy AI and sets a global benchmark," said a senior official at the European Commission.

Euronews coverage of EU rules on AI models become enforceable