Who needs to comply with the EU AI Act
Learn who must comply with the EU AI Act, from providers to deployers. Review the implementation timeline, key exemptions, and non-compliance penalties.
In this article8
- Scope and jurisdiction: Extraterritorial reach of the EU AI Act
- Providers versus deployers: Roles in the value chain
- General-purpose AI (GPAI): Compliance for foundational models
- Exemptions: National security, R&D, and open source
- Timeline and deadlines: When obligations take effect
- Penalties for non-compliance: The tiered fine structure
- Frequently asked questions
- Sources

Scope and jurisdiction: Extraterritorial reach of the EU AI Act
The EU Artificial Intelligence Act, Regulation (EU) 2024/1689, reaches well beyond the geographical boundaries of the Union. Under Article 2, applicability turns on market placement, operational location and downstream output usage, not on corporate headquarters[1]. Any foreign entity placing an AI system or GPAI model on the EU market, or putting it into service in the Union, falls under the Act wherever it is established[1].
The sharpest edge is the output-usage criterion in Article 2(1)(c). Providers and deployers established in a third country (the United States, the United Kingdom or Switzerland, say) must comply where the output produced by their AI system is used in the Union[1]. Protection therefore follows the European end-user, not the location of the compute, mirroring the extraterritorial scope seen in European data protection blockers.
- Providers placing AI systems or general-purpose AI (GPAI) models on the EU market, whether established in the EU or in a third country.
- Deployers of AI systems established or located in the European Union.
- Providers and deployers in third countries whose AI system outputs (decisions, predictions, generated content) are used in the Union.
- Importers and distributors placing AI systems into circulation across EU Member States.
- Product manufacturers integrating AI as a safety component under their own name or trademark.
The Act also protects affected persons in the Union subject to AI-assisted decisions, biometric evaluation or automated profiling[1]. Conversely, Article 2(10) exempts deployers who are natural persons using AI in a purely personal, non-professional activity[1].
Providers versus deployers: Roles in the value chain
Regulation (EU) 2024/1689 spreads obligations across distinct economic operators: Article 2(1) names providers, deployers, importers and distributors, product manufacturers and authorised representatives as the addressees[1]. Article 3 defines a provider as any person, public authority or agency that develops an AI system or GPAI model, or has one developed, and places it on the market or puts it into service under its own name or trademark[2]. A deployer is any entity using an AI system under its authority in a professional capacity, excluding purely personal, non-professional use[1].
| Operator Role | Statutory Definition | Core Regulatory Obligations | Key Articles |
|---|---|---|---|
| Provider | Develops an AI system or GPAI model and places it on the market under its own name | Risk management, data governance, conformity assessment, technical documentation, post-market monitoring | Article 16, Article 53 |
| Deployer | Uses an AI system under its authority in a professional activity (excluding private personal use) | Human oversight, performance monitoring, input-data relevance, automated log retention, fundamental rights impact assessment where required | Article 26 |
| Importer | EU-established entity placing an AI system from a third country on the Union market | Verifies conformity assessment, technical documentation, CE marking and authorised-representative designation | Article 23 |
| Distributor | Makes an AI system available on the EU market without affecting its properties | Verifies compliance markings and user instructions; ensures storage and transport do not compromise conformity | Article 24 |
Providers bear the primary compliance burden: technical documentation, continuous risk management, data quality standards, and CE marking before market entry[3]. Deployer obligations are operational: qualified personnel for human oversight, output monitoring, log retention for high-risk systems, and halting operations if an unexpected risk arises[5].
Article 25 adds a re-classification mechanism. If an importer, distributor, deployer or other third party substantially modifies a high-risk AI system, changes its intended purpose, or markets it under its own brand, that entity assumes the original provider's obligations[8]. For organizations evaluating specialized AI inference pipelines, that decides whether liability stays with the base model supplier or shifts to the deploying enterprise.
General-purpose AI (GPAI): Compliance for foundational models
Chapter V of Regulation (EU) 2024/1689 sets dedicated rules for general-purpose AI (GPAI) models. Under Article 53, every provider placing a GPAI model on the EU market must keep up-to-date technical documentation on training, testing and evaluation, make information available to downstream developers integrating the model, adopt a policy to comply with Union copyright law (including the Article 4(3) rights reservation in Directive (EU) 2019/790), and publish a sufficiently detailed summary of the training content on the AI Office template[4].
Systemic risk classification and cumulative compute thresholds
A further tier of obligations applies to GPAI models carrying systemic risk. Under Article 51, a GPAI model is classified as having systemic risk either through a formal European Commission decision or when the cumulative compute used for its training exceeds 10^25 floating-point operations (FLOPs)[9].
- Continuous model evaluation, including standardised adversarial testing (red-teaming).
- Assessing and mitigating systemic risk at Union level, including major cybersecurity breaches, autonomous capabilities and biological threats.
- Tracking, documenting and promptly reporting serious incidents and corrective measures to the AI Office and national authorities.
- Physical and cybersecurity safeguards for model weights, parameters and training infrastructure.
Article 56 establishes the GPAI Code of Practice, drawn up with independent experts, industry and civil society under the AI Office's coordination. Article 53(4) lets GPAI providers rely on such a code to demonstrate compliance with their transparency, documentation and copyright duties until a harmonised standard exists; a provider following neither route must show alternative adequate means for the Commission to assess[4].
Exemptions: National security, R&D, and open source
Article 2 of Regulation (EU) 2024/1689 also sets out targeted exclusions where the Act does not apply at all[1].
Defence, scientific research, and pre-market development
- Military and defence: under Article 2(3), the Act does not apply to AI placed on the market, put into service or used exclusively for military, defence or national security purposes, whatever the type of entity[1].
- Pure scientific research: Article 2(6) excludes AI systems and models, and their output, developed and put into service solely for scientific research and development[1].
- Pre-market R&D and testing: Article 2(8) excludes research, testing and development before an AI system is placed on the market or put into service, though testing in real-world conditions remains covered[1].
- International law enforcement cooperation: Article 2(4) excludes third-country public authorities and international organisations using AI under cooperation agreements for law enforcement and judicial cooperation, subject to adequate fundamental rights safeguards[1].
Conditional exemptions for open-source AI
Article 2(12) exempts AI systems released under free and open-source licences[1]. Article 53(2) adds that the GPAI documentation and downstream information duties do not apply to models under a free and open-source licence whose parameters, weights, architecture and usage information are public, though that exception falls away for models with systemic risk[4]. The exclusion is strictly conditional: it does not hold where such a system is placed on the market or put into service as high-risk, or falls under Article 5 or Article 50. Where such a model is rented rather than self-hosted, the exemption still turns on the licence and the Article 2(12) conditions, not on where inference runs.
Timeline and deadlines: When obligations take effect
Regulation (EU) 2024/1689 was published in the Official Journal on 12 July 2024 and entered into force on 1 August 2024. Its requirements then follow a staggered schedule, which decides not whether an organisation is in scope but when[10].
| Effective date | Statutory scope | What starts to apply |
|---|---|---|
| 2 February 2025 | Chapter I & Chapter II | Article 5 prohibitions and the Article 4 AI literacy duties start to apply. |
| 2 August 2025 | Chapter V & Chapter VII | Chapter V GPAI duties, Chapter VII governance and the Article 99 penalty tiers start to apply. |
| 2 August 2026 | General application | The rest of the Act applies, except the high-risk regime deferred by the AI Omnibus (rows below); the deadline for national AI regulatory sandboxes moved to 2 August 2027. |
| 2 December 2027 | Annex III (as amended) | Stand-alone high-risk AI systems in the Annex III areas, moved from 2 August 2026 by the AI Omnibus. |
| 2 August 2028 | Annex I & Article 111 | High-risk AI embedded as a safety component in products already covered by Annex I harmonisation law. |
Under the AI Omnibus (Regulation (EU) 2026/1744), co-legislators reached a political agreement on 7 May 2026 that moved the Annex III stand-alone high-risk date to 2 December 2027[10]. The deferral gives enterprises expanding their workplace AI adoption time to work against harmonised standards and sandboxes before high-risk duties bite.
Penalties for non-compliance: The tiered fine structure
Article 99 sets a graduated administrative penalty system that must be effective, proportionate and dissuasive[13]. For scoping it matters only as the size of the exposure a role carries.
| Infringement category | Statutory basis | Maximum administrative fine |
|---|---|---|
| Prohibited AI practices (Article 5) | Article 99(3) | Up to 35 000 000 EUR or 7 % of total worldwide annual turnover, whichever is higher |
| Breach of operator obligations or high-risk requirements | Article 99(4) | Up to 15 000 000 EUR or 3 %, whichever is higher |
| Supplying incorrect, incomplete or misleading information | Article 99(5) | Up to 7 500 000 EUR or 1 %, whichever is higher |
| Infringements by EU institutions and bodies | Article 100 | Up to 1 500 000 EUR for prohibited practices and 750 000 EUR otherwise, enforced by the EDPS |
Under Article 99(6), fines on SMEs and start-ups take whichever of the euro figure and the turnover percentage is lower[13]. For larger undertakings the ceiling is whichever is higher.
The final amount is set by the national competent authority against the mitigating and aggravating factors listed in Article 99(7)[13]. Understanding these jurisdictional boundaries and enforcement tiers is critical when designing and auditing European AI infrastructure deployments.
Frequently asked questions
Who is considered a provider under the EU AI Act?
A provider is any natural or legal person, public authority or agency that develops an AI system or GPAI model and places it on the market under its own name or trademark. Providers carry the heaviest burden, including conformity assessment and post-market monitoring.
Are open-source AI models exempt from the AI Act?
AI systems released under free and open-source licences are out of scope unless placed on the market or put into service as high-risk systems, or falling under Article 5 or Article 50.
Does the EU AI Act apply to companies outside of Europe?
Yes. It applies to providers and deployers established in third countries where the output produced by their AI system is used in the Union.
When do the rules for high-risk AI systems take effect?
Following the AI Omnibus amendment, Annex III high-risk systems take effect on 2 December 2027. High-risk AI embedded in regulated products, such as medical devices or lifts, has until 2 August 2028.
What are the penalties for non-compliance with the EU AI Act?
Article 99 sets three tiers: up to 35,000,000 EUR or 7% of total worldwide annual turnover for prohibited practices, 15,000,000 EUR or 3% for breaches of operator obligations, and 7,500,000 EUR or 1% for incorrect or misleading information.
Are military and national security applications covered?
No. The Act does not apply to AI placed on the market, put into service or used exclusively for military, defence or national security purposes, whatever the type of entity.
Sources
- Article 2: Scope ai-act-service-desk.ec.europa.eu
- Article 3: Definitions ai-act-service-desk.ec.europa.eu
- Article 16: Obligations of providers of high-risk AI systems ai-act-service-desk.ec.europa.eu
- Article 53: Obligations for providers of general-purpose AI models ai-act-service-desk.ec.europa.eu
- Article 26: Obligations of deployers of high-risk AI systems ai-act-service-desk.ec.europa.eu
- Article 23: Obligations of importers ai-act-service-desk.ec.europa.eu
- Article 24: Obligations of distributors ai-act-service-desk.ec.europa.eu
- Article 25: Responsibilities along the AI value chain ai-act-service-desk.ec.europa.eu
- Article 51: Classification of general-purpose AI models as general-purpose AI models with systemic risk ai-act-service-desk.ec.europa.eu
- AI Act digital-strategy.ec.europa.eu
- Implementation Timeline artificialintelligenceact.eu
- consilium.europa.eu
- Article 99: Penalties ai-act-service-desk.ec.europa.eu
- Article 100: Administrative fines on Union institutions, bodies, offices and agencies ai-act-service-desk.ec.europa.eu
Rankings and scores on Optinest cannot be bought — no commercial relationship changes how a provider is assessed. How we score
Related reading
All Regulation →What the AI Act asks of your business, and when
Understand how the EU AI Act impacts your business. Learn the distinct obligations for AI providers and deployers, timelines, and data sovereignty rules.

LLM APIs with zero data retention: A provider comparison
Compare 7 LLM inference providers that offer zero data retention. Learn what each commitment covers and how to get API logging restrictions in writing.

Renting 2- and 4-GPU servers instead of a full node
Compare European providers that rent 1, 2 or 4 GPUs instead of a full 8-GPU node, with per-GPU rates from each pricing page and the NVLink trade-off explained.
