A date that needs two meanings
The EU AI Act entered into force on 1 August 2024. [3] That historical event matters, but it should not be retold as the day every obligation became immediately applicable to every company, public body, model provider or deployer. Entry into force made the Regulation part of EU law. Applicability determines when particular provisions begin to operate. The distinction is technical, but it changes how the August 2024 moment should be understood.
The Commission’s announcement presented the Act as a uniform EU framework built around risk. It described minimal-risk systems, transparency-risk systems, high-risk systems and unacceptable-risk practices. It also said the Act addressed potential risks to health, safety and fundamental rights, while setting requirements for developers and deployers. Those statements are the Commission’s official description of policy purpose and framework. They are not independent evidence that every obligation would be easy to apply or that the regime would produce its intended outcomes.
This archive edition therefore treats August 2024 as a legal starting point, not a retrospective announcement of instant, complete compliance. That framing is especially important for general-purpose AI, whose uses can extend through many products, organisations and sectors.
The original policy context
The speed with which general-purpose and generative AI became prominent sharpened the regulatory debate. The academic analysis published on 1 August 2024 characterized the Act’s approach to general-purpose AI as a significant shift from reactive to proactive governance. [1] Yet its conclusion was qualified: enforceability, democratic legitimacy and future-proofing remained live questions. The authors argued that success hinged on effective enforcement, productive intra-European and international cooperation, and the EU’s ability to adapt to a rapidly evolving AI landscape.
That is a more disciplined reading than declaring the Act either a solved governance problem or an empty symbolic gesture. Rules can create duties, institutions and common vocabulary. They do not automatically settle how authorities will assess complex systems, how actors across a value chain will share information, or how requirements will remain meaningful as technology changes.
The same analysis discussed risks associated with general-purpose AI including unreliability and lack of transparency, cyber-security concerns, and systemic effects connected to concentration of power. These are analytical risk categories, not findings that all models or deployments cause each harm. The distinction matters operationally: risk-based governance requires attention to a specific system and use context, not assumptions based solely on a model label.
The phased timetable
Article 113 is the essential corrective to the shorthand that “the Act started in August 2024.” It says the Regulation applies from 2 August 2026, while setting earlier application dates for specified provisions. Chapters I and II apply from 2 February 2025. Specified provisions, including Chapter V, apply from 2 August 2025; Article 6(1) and corresponding obligations apply from 2 August 2027. [2]
The timeline creates three separate reference points. First came entry into force on 1 August 2024. Next came staged applicability beginning in 2025. General application followed on 2 August 2026, subject to the exceptions stated in Article 113. A credible historical account must preserve those differences rather than collapse them into a single deadline.
For organisations in 2024, the transition period was not regulatory emptiness. It was time to establish an AI inventory; identify whether the organisation was acting as provider, deployer, purchaser, integrator or another actor; map suppliers and downstream uses; and establish routes for documentation, escalation and oversight. These are practical interpretations of a phased regulatory design, not additional legal duties asserted by the supplied sources.
What later material adds—and does not change
Later Commission material must not be projected backward onto the August 2024 event. It says GPAI rules became effective in August 2025. It also says that, from 2 August 2026, the AI Office and Member State authorities are responsible for implementation, supervision and enforcement, and that the AI Office has enforcement powers over GPAI models. [4]
These are later or current institutional facts in the supplied archive. They are not evidence that this enforcement arrangement was already operating on 1 August 2024. The sequence instead shows an architecture moving from legal adoption, through early provisions and GPAI rules, into wider application and enforcement.
Current Commission material describes GPAI rules concerning transparency and copyright, and says providers of GPAI models that may pose systemic risks should assess and mitigate those risks. This is an official account of the framework. It should not be mistaken for independent proof that risk assessments detect every significant issue or that mitigation is always effective.
Practical reading for leaders
The useful management question is not “Were we covered in August 2024?” It is: which role, system, use case and date apply to us now? A sound programme maps each AI-enabled workflow to its purpose, supplier, affected users, risk context, applicable timeline, controls and evidence. It distinguishes a general-purpose model from the application built on top of it, because deployment context can alter transparency needs, human oversight and relevant risks.
Procurement is part of that work. Buyers should seek contractual access to documentation, information on material changes, escalation paths and clear responsibilities. Technical teams should preserve records that explain intended purpose, operational limits, monitoring and decisions about human review. These measures do not guarantee compliance or safety. They make an organisation better able to identify what it operates and substantiate how it governs it.
Limits and lesson
This article relies solely on the supplied captures. The 2024 academic source is an initial interdisciplinary analysis, not an enforcement evaluation. Commission pages are authoritative statements of the EU framework, not independent effectiveness studies. Later material is labelled as later or current rather than used to rewrite the August 2024 record.
The enduring lesson is narrow but consequential: entry into force established the Act in law; phased applicability determined when specified obligations applied. [3] [2] Effective outcomes remain contingent on enforcement, cooperation and adaptation as well as operational preparation. [1]