Regulation (EU) 2024/1689 · the EU AI Act
The EU AI Act is here. Find out what you owe — and by when.
The world’s first comprehensive AI law sorts every system into four risk tiers. Some uses are already banned; high-risk duties land in 2027. This page decodes it by tier, by date, and by article, in plain language.
Unacceptable
Banned outright · Art. 5
In force Feb 2025 · €35M / 7%
High risk
Full obligations · Arts. 9–15, 19
From Dec 2027 · €15M / 3%
Limited risk
Transparency only · Art. 50
From Aug 2026
Minimal risk
No obligations · voluntary codes
Most AI today
How the Act sorts every AI system
Higher tier = more obligation. + GPAI overlay · Arts. 51–56.
Does this apply to me?
Find your obligation in one line.
The Act regulates by what your AI does, not what industry you’re in. Match your case below to see the tier, the deadline, and the maximum fine.
Article 113 · staggered application
Two deadlines have passed. Three are ahead.
Each date switches on a wall of obligations. The next one is highlighted.
2 Feb 2025
Bans + AI literacy
Art. 4 & Art. 5 apply.
2 Aug 2025
GPAI + governance
Art. 53 duties and penalty applicability begin.
2 Aug 2026 · next
GPAI fines attach
Commission fines under Art. 101 become exercisable.
2 Dec 2027
High-risk wave
Annex III duties (Arts. 6–15, 19). Deferred by the Digital Omnibus.
2 Aug 2028
Sectoral high-risk
Annex I product-embedded systems.
Articles 99 + 101
Four fine bands. One CFO conversation.
Article 5 prohibitions carry the top band, €35M or 7% of global turnover, whichever is higher. Bars drawn to scale.
Article 99(6) · SME proportionality. For SMEs and start-ups, the fine is whichever is lower of the turnover percentage or the euro cap, the inverse of the rule for larger firms.
Article browser
Pick an article. See exactly what it asks of you.
Plain-language summary, who it hits, enforcement date, penalty band, and the tool that produces the evidence, in one click.
Source: Regulation (EU) 2024/1689 · Official Journal of the European Union · Not legal advice.
Cross-cutting explainers
Two scope regimes. Different rulebooks.
Article 6 + Annex III
What counts as high-risk
Eight Annex III use-cases trigger the full obligations stack (risk management, data governance, documentation, accuracy). Profiling keeps a system in scope.
Articles 51–56
GPAI model providers
General-purpose AI sits on a separate chain, enforced by the European Commission via the AI Office, not member states. A systemic-risk overlay activates above 10²⁵ FLOPs.
A 7% turnover penalty is a number a CFO sees before any technical detail lands.
Our tools produce the evidence; turning it into a programme, a board narrative and regulator dialogue is human work. That is the advisory arm.