There is no single application date
The AI Act is the European regulation that establishes obligations for artificial intelligence systems based on risks and the roles of operators. It entered into force on August 1, 2024, but its provisions become applicable in phases. Speaking generically of the “AI Act date” therefore risks confusing entry into force, application of individual articles, and deadlines for different categories of systems.
The European Commission’s official timeline sets out successive milestones: the prohibitions and provisions on definitions and AI literacy from February 2, 2025; obligations for general-purpose AI models and certain governance measures from August 2, 2025; most rules from August 2, 2026. From December 2, 2026, new prohibitions also apply to systems that generate non-consensual sexual deepfakes or child sexual abuse material, together with a specific transition for certain providers of generative systems already placed on the market before August 2026. Additional dates are envisaged for certain categories of high-risk systems, including December 2, 2027, and August 2, 2028; from August 2, 2027, Member States should also have at least one operational regulatory sandbox. The updated timeline takes into account the amendments introduced by the Digital Omnibus on AI: dates and obligations should therefore be rechecked on the official page.
The timetable alone does not establish obligations
To understand which provision applies, it is necessary to identify the type of system, the intended purpose, the organization’s role, and the date of placing on the market or putting into service. A general-purpose model provider, a deployer using a system at work, and an importer may have different duties. High-risk systems also depend on the sector and conditions of use: the commercial label “AI” is not enough to classify a product.
The timeline also distinguishes between obligations for general-purpose models, high-risk systems, transparency, and governance. Some rules concern the manufacturer or provider; others fall on those who integrate or use the system. A deadline may therefore be relevant to one actor and not to another.
How to Read News About Deadlines
When an article announces that “the AI Act enters into force,” check whether it refers to the original date of the regulation or to the application of a specific rule. Look for the category concerned, the role of the actor, and the link to the updated official text. If a news report cites a postponement or an amendment, check whether it is a proposal, a political agreement, or a definitively adopted rule.
The Commission makes available a timeline and orientation materials, but these do not replace the legal text or an analysis of the specific case. Updates may change timing and procedures; for an organization that must comply, it is prudent to keep the version consulted and to verify the guidance of the competent authority.
Three Essential Checks
Does the news report distinguish entry into force from the application of individual provisions? Does it specify which category of system and which role are involved? Does it link to an updated institutional page instead of repeating an outdated table? These questions help avoid hasty conclusions.
This guide offers general journalistic orientation on the dates reported by the Commission, not a legal opinion or a compliance assessment. For operational decisions, it is necessary to consult the updated text and verify how it applies to one’s own system.
