[ policy.watch ] · European Union
European AI policy
What is changing for the people building, buying and using AI in the EU.
Sources checked 18 November 2026
The latest changes
ADOPTED AMENDMENTS
AI Omnibus
The Commission reports that the AI Omnibus entered into force on 27 July 2026. It extends high-risk application dates, expands sandbox access, extends some SME support to small mid-cap companies and simplifies administrative obligations. It is no longer described here as merely a proposal.
APPLICABLE RULES + GUIDANCE
Transparency in AI interactions
Article 50 applies from 2 August 2026. The Commission’s guidance describes informing people about direct AI interactions, machine-readable marking of synthetic content, and disclosures for certain deepfakes and public-interest text. Scope, exceptions and transitional provisions matter; not every AI-assisted text needs the same label.
Implementation timeline
Status as of 18 November 2026; future dates are the official published milestones, not guarantees against further amendments.
Milestone reached
Initial prohibitions and general provisions
The initial AI Act prohibitions and general provisions began applying. The 2026 Omnibus subsequently simplified the AI-literacy requirement.
Milestone reached
General-purpose AI model rules
Rules for providers of general-purpose AI models began applying, alongside governance requirements. Transitional provisions can affect particular models.
Milestone reached
AI Omnibus enters into force
The Commission reports adopted amendments, including extended high-risk timelines and simplified requirements. Entry into force does not mean every amended provision applies immediately.
Milestone reached
Transparency obligations apply
Article 50 transparency obligations began applying to relevant AI systems. The official timeline identifies a later transition for certain pre-existing synthetic-content systems.
Upcoming
New prohibitions and synthetic-content transition
The official timeline sets this date for prohibitions on non-consensual sexual deepfakes and child sexual abuse material, and an Article 50(2) transition for certain systems placed on the market before 2 August 2026.
Upcoming
National AI regulatory sandboxes
Member States should have at least one AI regulatory sandbox operational per country.
Upcoming
Annex III high-risk systems
The amended timeline sets this application date for rules covering high-risk AI systems in Annex III.
Upcoming
High-risk AI in regulated products
The amended timeline sets this date for high-risk AI embedded in regulated products covered by Annex I.
What to check before using an AI tool
- Identify whether you are a provider, deployer or model provider; the rules differ.
- Assess the actual use case and risk category, especially in employment, education or other consequential decisions.
- Ask your vendor about disclosures, documentation, human oversight and data handling.
- Do not treat European headquarters, funding or an “EU-owned” label as proof of legal compliance.