Check which rules may apply
The Commission's official Compliance Checker is in beta and helps providers, deployers and other operators identify possible obligations. Its result is guidance, not a legal determination.
Open the Compliance Checker →Updated 26 September 2026
AI and the rules
Most explainers online were written before 27 July 2026 and run on a timeline that no longer holds. Here are the dates as they stand, each linked to the legal source.
This is information, not legal advice. Every rule below is cited to its source and reflects the position on 26 September 2026. This area is moving quickly, and a decision that matters for your organisation should be checked by someone with legal responsibility for it.
The short version
On 27 July 2026 the EU brought an amending regulation into force that pushes the heavy obligations for high-risk AI back by roughly eighteen months. The transparency obligations were left exactly where they were. The result is that a lot of organisations are preparing for a deadline that moved, while missing the one that did not.
If you use AI for writing, analysis, customer contact or images, three questions actually matter: do you publish AI-generated content, have you given staff any training, and what personal data goes into the tools.
Timeline
What applies when
Taken from the regulation text and the European Commission's own pages. Scroll sideways on a small screen.
ProvisionDateStatus
Prohibited AI practices (Art. 5)2 February 2025In force
AI literacy duty for staff (Art. 4)2 February 2025In force
General-purpose AI model rules2 August 2025In force
Transparency and content marking (Art. 50)2 August 2026In force
Machine-readable marking, existing systems2 December 2026Transition
Stand-alone high-risk systems (Annex III)2 December 2027Delayed from 2026
High-risk AI inside regulated products (Annex I)2 August 2028Delayed from 2027
New from the EU
Official help, a compliance checker and a complaints channel
The Commission now provides three practical routes that were missing from many of the early explainers.
Ask the AI Office
The AI Act Service Desk accepts questions in every official EU language. An EU Login is required, and the reply is delivered by email.
Open the Service Desk →Complain about a possible breach
People and organisations can submit an Article 85 complaint. It is not anonymous and only covers matters within the AI Office's exclusive competence — not other EU law or national law.
Read about the complaints tool →Scope matters: this is not a general route for every problem involving AI. The AI Office may refer an eligible complaint to the appropriate national authority with the complainant's consent. Eligible people professionally connected to an AI provider or deployer may instead use the Commission's separate anonymous whistleblower tool.
The delay that changed the picture
The transparency rules took effect on 2 August 2026, and both the Commission and the national market surveillance authorities began enforcing the regulation on the same day. From that date the AI Office can also fine providers of general-purpose AI models.
The amending regulation, known as the Digital Omnibus, was published in the Official Journal on 24 July 2026 and entered into force on 27 July 2026. It moves the deadline for stand-alone high-risk systems in Annex III from 2 August 2026 to 2 December 2027, and for high-risk AI embedded in already-regulated products from 2 August 2027 to 2 August 2028.
So the delay covers the categories that hit recruitment, worker management, education, credit scoring and access to essential services. It does not cover the transparency rules, and it does not touch the prohibitions.
Myth
“2 August 2026 is the big deadline when everything applies.”
What is actually trueIt is the date for transparency and content marking. The heavy high-risk requirements moved to December 2027.
Myth
“This only applies to companies that build AI models.”
What is actually trueSeveral duties land on whoever puts the system to use. Marking content and training staff are both deployer obligations.
Myth
“We only use ChatGPT internally, so we are out of scope.”
What is actually trueInternal use rarely triggers high-risk duties, but the AI literacy duty still applies, and data protection law applies from the first thing you paste in.
Go deeper
The questions that actually reach you
Which category are you in?
Six questions that place your AI use, and say when the rules apply.
Read the guide →Labelling AI content
Chatbots, deepfakes and published AI text. Applies from 2 August 2026.
Read the guide →AI literacy for staff
Required since February 2025, and softened on 27 July 2026.
Read the guide →Editorial responsibility
The one place the law asks less of you when a human genuinely reviews the work.
Read the guide →Personal data in AI tools
Data protection law applies regardless of what happens to the AI Act.
Read the guide →High-risk uses at work
Recruitment, performance management and the two outright bans.
Read the guide →Who was first?
The EU was first with a horizontal AI law. On labelling, China beat it by eleven months.
Read the guide →Sources
- Regulation (EU) 2026/1744, the amending regulation — published 24 July 2026, in force 27 July 2026
- European Commission: the AI omnibus enters into force — the Commission's own summary
- Commission FAQ on Article 50 — the transparency duties in detail
- Regulation (EU) 2024/1689, the AI Act — the base regulation
- AI Act Service Desk — the Commission's article-by-article reference
- European Commission AI Act complaints tool — scope, identification and complaint handling
- Official AI Act Compliance Checker — a beta tool for an initial view of possible obligations