News · 13 August 2024
The AI Act is in force: a timeline for people who build and buy AI
The EU AI Act entered into force on 1 August 2024. That does not mean every rule applies at once. The regulation phases in, and the dates matter for planning.
The staged timeline (as set out in 2024)
- Six months after entry into force: the prohibited practices and the general provisions, including AI literacy, apply.
- Twelve months: rules for general-purpose AI models, along with governance structures and penalties.
- Twenty-four months: most remaining obligations, including those for high-risk systems listed in the annexes.
- Thirty-six months: obligations for high-risk AI that is a safety component of products already covered by other EU product legislation, such as medical devices.
For healthcare developers the last point is the one to underline. AI in a regulated medical device sits in a longer timeline than a standalone high-risk use, but it must also satisfy medical device rules, so the work is cumulative rather than alternative.
Who is affected?
The Act applies to providers who place AI on the EU market and deployers who use it, including organisations outside the EU whose systems' outputs are used there. A UK company supplying a model to a hospital in Italy is in scope. So is an Italian hospital deploying a UK-built tool.
What to do with a two-year runway
- Inventory your AI systems and classify the intended use of each.
- Start documentation now. Technical documentation, data lineage and validation records are far easier to build as you go than to reconstruct later.
- Plan for logging and oversight. Decide what you will record and who can intervene.
- Treat explainability as a requirement. Interpretability supports the transparency and oversight duties that high-risk rules rely on.
A note on proportion
Most AI uses will not be high-risk. The goal of classification is to put effort where it matters, not to burden every spreadsheet model. If a use is not high-risk, the sensible response is still to apply light-touch versions of the same habits, because they make systems better.
Written in August 2024; details reflect the situation at that time.
Update, October 2026: the EU's Digital Omnibus on AI has since postponed the application dates for high-risk systems. We cover the revised dates in later articles.