AI law and policy in European Union
8 instruments relevant to AI in European Union are indexed here, 6 of them binding law, 2 of which name AI or deepfakes in their titles.
Use of AI, AI datacentres and governance readiness beside this index: AI in the European Union
Statutes
Act of Parliament. Binding on everyone it applies to.
-
European Parliament and Council
Amends the AI Act, in force from July 2026, six days before the high-risk obligations were due to bite. Stand-alone high-risk systems move to December 2027 and systems embedded in regulated products to August 2028. The Article 50 transparency duties were not delayed. Treat any compliance date published before July 2026 as out of date.
- safety
- transparency
-
European Parliament and Council
Brings software and AI systems inside the product liability regime and eases the burden of proof where a claimant could not reasonably be expected to explain how a complex system failed. The separate AI Liability Directive was withdrawn in 2025, so this is what remains.
- liability
- consumer
- safety
-
European Parliament and Council
The first horizontal AI statute to enter into force. Bans a short list of practices, imposes conformity assessment on high-risk uses, sets transparency duties on systems people interact with, and adds obligations for general-purpose models above 1e25 FLOP of training compute. It applies in stages rather than at once, and those stages have since moved.
- safety
- transparency
- labelling
- consumer
-
European Parliament and Council
Systemic-risk assessment and mitigation duties on very large platforms and search engines, explicitly including risks from recommender and generative features. Where a model ships inside a large platform, this usually binds before the AI Act does.
- safety
- transparency
- elections
-
European Parliament and Council
Article 4 permits text and data mining of lawfully accessible works unless the rightsholder reserved the right in a machine-readable way. That reservation is the legal basis for most European objections to training on web data.
- copyright
-
European Parliament and Council
Predates the AI Act and reaches further in practice. Article 22 restricts decisions based solely on automated processing that produce legal or similarly significant effects, and it is the provision most European automated-decision disputes are actually fought under.
- privacy
- transparency
Binding on government only
Binds government agencies only, not the wider market.
-
European AI Office 2024
European Commission
Supervises general-purpose model obligations directly rather than leaving them to national authorities. It is the only body anywhere with statutory supervisory powers aimed specifically at frontier models.
- safety
- government use
Voluntary standards
Guidance with no legal force. Compliance is a choice.
-
European Commission and independent experts
Voluntary, but signing it is the route to presumed conformity with the AI Act's general-purpose model obligations. A voluntary instrument with that consequence sits between guidance and regulation, which is why it is indexed.
- safety
- transparency
- copyright
About this index
Compiled here from primary instruments and licensed MIT.
6 of the 8 instruments here are binding law, and 2 of those name AI. The remainder apply to AI without naming it, so a keyword search of the statute book does not find them.
Topics covered: consumer, copyright, elections, government use, labelling, liability, privacy, safety, transparency. A build check probes every citation on this page, because a link that returns a 404 looks like evidence and is not.
This is not legal advice. It is not a compliance checklist and it is not exhaustive. Commencement dates, proposals and enforcement positions move; the review date is on every record. If something here is wrong, say so and it gets fixed and listed rather than quietly patched.