A.I.
A.I. · Reference
AI regulation
The EU AI Act is the European Union's law on artificial intelligence and the most complete AI law in force. It sorts AI by risk: some uses are banned, high-risk uses face strict checks, and general-purpose models, the big models labs ship, have their own duties. This page tracks the dates that matter.
Key dates
| Date | What applies |
|---|---|
| 2 August 2025 | Duties for general-purpose AI model makers: technical documents, a copyright policy, and a public summary of training data. Bigger models with "systemic risk" must also test for and report serious risks. |
| 2 August 2026 | The EU AI Office and national authorities start enforcing the Act. New transparency rules apply: chatbots must say they are AI, deepfakes must be labelled, and AI-made content must carry machine-readable marks. |
| 27 July 2026 | Latest consolidated text of the Act on EUR-Lex, after amendments. |
Sources: the European Commission's 31 July 2026 release (Commission), its questions and answers on general-purpose models (Commission Q&A), and the consolidated law (EUR-Lex). A deepfake is an image, video, or audio clip made or changed by AI to look real. Machine-readable marks are hidden signals, like watermarks, that software can detect.
Why it matters for this site
- Model makers publish training-data summaries because of this law. Cohere's Command A+ card links its EU AI Act Article 53 summary.
- Meta's Content Seal watermark on Muse Image is the kind of machine-readable mark the transparency rules ask for.
Not covered yet
This page does not yet track US federal or state AI laws, or the later EU deadlines for high-risk systems, which were changed by amendments in 2026. Those need primary sources before they go up.