When do EU AI Act obligations start applying?

Short conclusion

Regulation (EU) 2024/1689 (the "AI Act") entered into force on 1 August 2024, but most of its obligations only apply from 2 August 2026. A small set of provisions applies earlier, and one set is delayed further. There are no exceptions to those dates in the typical case; the only flexibility is a narrow, delegated power for the European Commission to defer certain deadlines for a limited set of entities.

---

The general rule: 2 August 2026

Under Article 113(1) of the AI Act, the Regulation as a whole applies from 2 August 2026. For the typical provider, deployer, importer, distributor or producer of an AI system placed on the Union market, this is the date on which the substantive obligations (conformity assessment, registration, post-market monitoring, fundamental rights impact assessment for certain deployers, transparency duties, etc.) become binding. The24-month delay was built in to give operators, notified bodies and national competent authorities time to prepare.

Earlier application:2 February 2025

Article 113(2)(a) brings forward two whole chapters:

• Chapter I (Articles 1–4): scope, definitions, general principles and AI literacy obligations under Article 4. From2 February 2025, providers and deployers must already ensure a sufficient level of AI literacy among their staff and others operating or using AI systems on their behalf.
• Chapter II (Article 5): the prohibited AI practices — for example, subliminal manipulation, exploiting vulnerabilities of specific groups, social scoring by public authorities, real-time remote biometric identification in publicly accessible spaces for law enforcement (subject to narrow exceptions), and emotion recognition in workplaces and educational settings.

From that date, putting a prohibited system on the market or into service is itself unlawful, even though the rest of the Act is not yet in force.

Earlier application: 2 August 2025

Article 113(2)(b) advances:

• Chapter V, Section 4 (Articles 51–55): obligations for general-purpose AI (GPAI) models and, from the designation made by the Commission, for GPAI models with systemic risk. This is the date providers of foundation models must comply with transparency, copyright and (for systemic-risk models) risk-assessment and incident-reporting duties.
• Article 85 (national competent authorities), Article 91 (mutual assistance, market surveillance), Article 94 (penalties for GPAI providers) and Article 95 (general penalties for undertakings).

Later application: 2 August 2027

Article 113(3) defers one category: the obligations of providers and deployers of high-risk AI systems listed in Annex III (biometrics, critical infrastructure, education and vocational training, employment, access to essential private and public services, law enforcement, migration and border control, administration of justice and democratic processes). For those systems, the2 August 2026 date is replaced by 2 August 2027, giving an additional year for compliance.

High-risk AI systems that are safety components of products already regulated by Union harmonisation legislation listed in Annex I follow the general date of 2 August 2026.

Main exceptions and flexibility

• Article 113(2): the two earlier dates above are themselves exceptions within Article 113. Outside them, the text of the Act does not create case-by-case exemptions from the timeline.
• Article 113(4) and (5): where a Member State has not yet designated the national competent authority or notified body by 2 August 2026, GPAI providers can rely on a lighter interim regime and the Commission's AI Office acts as the competent authority until the Member State acts. This is an implementation safeguard, not a delay of the substantive obligations.
• Article 113(6) and (7): for AI systems that are safety components of products covered by certain Annex I legislation (e.g. medical devices, machinery, motor vehicles), the AI Act applies only from the date the relevant sectoral regime itself becomes applicable. In other words, the AI Act synchronises with existing product-regulation timetables.
• Article 113a/Article 113(8) (Commission delegated power): the Commission may adopt delegated acts to defer the application dates by up to six months (to 2 February 2027 for the general date, 2 February 2028 for Annex III), and by up to two years specifically for high-risk AI systems listed in Annex III, in particular where providers encounter critical compliance bottlenecks. Exercising this power is exceptional and would be published in the Official Journal.

Practical takeaway for operators

If you place an AI system on the Union market in the ordinary course of business, plan on 2 August 2026 as your compliance date, with 2 February 2025 for the literacy duty and the prohibition list already in force.

---

Sources

The web research tools available for this answer did not return results for queries on Regulation (EU) 2024/1689, so the standard EUR-Lex reference links for the regulation and its articles cannot be quoted as retrieved. The standard CELEX identifier for the regulation is CELEX:32024R1689 on EUR-Lex (https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689); Article 113 is the provision that sets out the dates of application described above.

This article is general information, not legal advice. The exact compliance dates for a specific AI system (in particular whether it is high-risk under Annex I or Annex III, whether it is a GPAI model, or whether it falls under sectoral Annex I product legislation) depend on facts that need to be assessed case by case.