The EU AI Act enters into force

The European Commission had proposed an AI regulation in April 2021, before ChatGPT, built around risk categories for uses: some banned, such as social scoring and most real-time facial recognition by police; some high-risk, such as hiring, credit and medical devices, with obligations for testing and oversight; the rest lightly touched. The arrival of general-purpose models forced a late addition, and the version the Parliament adopted on 13 March 2024 added obligations for foundation models, with extra ones for those trained above 10²⁵ floating-point operations, deemed to carry systemic risk. The Act was published in July and entered into force on 1 August 2024.

It applied in stages. The prohibitions took effect in February 2025 and the general-purpose model rules, with a code of practice the major laboratories signed, in August 2025. The high-risk obligations were due in August 2026, until the Digital Omnibus of July 2026 pushed them to December 2027 and August 2028 under industry pressure.

The Act is on this timeline as the first law to define frontier models by the compute used to train them, and as the reference point for every other jurisdiction's debate about whether to regulate the technology at all.