Opens in a new tab

Who does the EU AI Act apply to? Scope, territorial reach and exemptions

Maarten Stolk

Maarten Stolk · 9 min read · Last reviewed 20 Oct 2026

In short

The EU AI Act applies to organisations that place AI systems or general-purpose AI models on the EU market, put AI systems into service in the EU, use AI systems in the EU, or run AI systems whose output is used in the EU. Where the organisation is established does not matter. Article 2 excludes a short list: military, defence and national security systems, AI built solely for scientific research, research and testing before market launch, purely personal use, and most free and open-source AI. Being in scope does not mean every obligation applies. What you owe depends on your role and on the system’s risk tier. The high-risk duties now start on 2 December 2027 for Annex III systems and 2 August 2028 for Annex I systems.

Questions that decide whether the Act applies

Article 2 comes down to four questions, asked in this order.

  1. Is the technology an AI system or a general-purpose AI model?
  2. What role do you play: provider, deployer, importer, distributor or manufacturer?
  3. What connects you to the EU?
  4. Does one of the Article 2 exclusions apply?

If the answers put you in scope, the next step is your risk tier. The risk classification guide and the Self-Assessment Tool cover that.

Question 1: is it an AI system?

The Act regulates systems that meet the definition in Article 3(1): a machine-based system designed to operate with varying levels of autonomy, which may exhibit adaptiveness after deployment, and which, for explicit or implicit objectives, infers from the input it receives how to generate outputs such as predictions, content, recommendations or decisions that can influence physical or virtual environments. The Commission’s guidelines on the definition, published on 6 February 2025, break it into seven elements and say it should not be applied mechanically. The guidelines are not binding.

They describe four groups of systems that fall outside the definition: systems for improving mathematical optimisation, basic data processing, systems based on classical heuristics, and simple prediction systems. The optimisation group refers to established methods such as linear and logistic regression. That matters for banks and insurers with older scorecards. The guidelines also say each system is assessed on its own characteristics, so a scorecard needs a documented assessment against the seven elements rather than an assumption either way.

General-purpose AI models are covered separately in Chapter V. A provider that places one on the EU market is in scope under Article 2(1)(a) wherever it is based. The guide on building on GPAI models covers the deployer side.

Question 2: what role do you have?

Article 2(1) lists the operators the Act binds, plus the people it protects.

RoleWho it coversArticle 2(1)
ProviderDevelops an AI system or general-purpose AI model, or has one developed, and places it on the market or puts it into service under its own name or trademark, for payment or free of chargePoints (a) and (c)
DeployerUses an AI system under its authority, other than in a purely personal non-professional activityPoints (b) and (c)
ImporterLocated or established in the EU and places on the market an AI system that bears the name or trademark of a person established in a third countryPoint (d)
DistributorIn the supply chain, other than the provider or the importer, and makes an AI system available on the EU marketPoint (d)
Product manufacturerPlaces an AI system on the market or puts it into service together with its product and under its own name or trademarkPoint (e)
Authorised representativeActs for a provider that is not established in the EUPoint (f)
Affected personLocated in the EUPoint (g)

Building a system for your own use makes you its provider. The Act defines “putting into service” as supplying a system for first use directly to the deployer or for own use in the Union for its intended purpose (Article 3(11)). A bank that builds a credit model in-house and runs it internally has therefore put an AI system into service. There is no internal-use exemption. A deployer can also become a provider, for example by customising a system or changing its intended purpose. The guide on Article 25 shows where that line sits.

Question 3: what connects you to the EU?

Three links bring an organisation into scope, and none of them depends on where it is established.

SituationResultArticle 2(1)
A provider, established anywhere, places an AI system or general-purpose AI model on the EU marketIn scopePoint (a)
A deployer is established or located in the EUIn scopePoint (b)
A provider or deployer is established or located outside the EU, and the system’s output is used in the EUIn scopePoint (c)

The output trigger in point (c) means that running a system outside the EU does not by itself take it out of scope when the results are used in the EU.

Question 4: which exclusions apply?

Article 2 contains a short list of exclusions. Each one has a condition that limits it.

ExclusionArticleThe condition that matters
Areas outside Union law2(3)Member States keep their competences on national security, whatever type of entity carries out the tasks
Military, defence and national security2(3)Only where the system is placed on the market, put into service or used, with or without modification, exclusively for those purposes
Public authorities of third countries and international organisations2(4)Only when they use AI in international cooperation or agreements for law enforcement and judicial cooperation with the Union or Member States, and the third country provides adequate safeguards for fundamental rights
Scientific research and development2(6)Only AI systems or models, output included, specifically developed and put into service for the sole purpose of scientific research and development
Research, testing and development before market launch2(8)Covers activity before the system is placed on the market or put into service. Testing in real-world conditions is not covered. Union law still applies
Personal, non-professional use2(10)Covers obligations of deployers who are natural persons using AI in a purely personal non-professional activity
Free and open-source licences2(12)Not excluded if the system is placed on the market or put into service as high-risk, or falls under Article 5 or Article 50

Read the conditions literally. “Exclusively” in the military clause and “sole purpose” in the research clause do the work. A system built for defence but also used for civilian purposes is not covered by that exclusion, and a research model that is later put into service for other purposes is not covered by the research one.

The research boundary comes up in regulated industries more than the military one. A data science team prototyping a claims model is outside the Act under Article 2(8). When the insurer puts that model into production for its own use, it has put an AI system into service and the exclusion no longer applies.

Two more rules limit how the Act reaches certain high-risk products:

  • For high-risk AI in products covered by the sector laws in Annex I, Section B, only Article 6(1), Article 60a and Articles 102 to 112 apply. Section B covers civil aviation, motor vehicles, two- or three-wheel vehicles, agricultural and forestry vehicles, marine equipment, rail interoperability and, since the Digital Omnibus, machinery. The sandbox articles (57 to 59) apply only so far as the Act’s requirements have been built into that sector legislation.
  • A new Article 2(13), added by the Digital Omnibus, allows the requirements in Articles 9 to 15 and 17 to 25 to be limited for high-risk AI in products under Annex I, Section A, which includes medical devices and radio equipment, where that sector law gives an equivalent or higher level of protection and the overall level of protection is not reduced. The Commission has to specify the systems, requirements and conditions through delegated acts by 2 August 2027.

The Act also sits beside other law rather than above it. It does not affect consumer protection and product safety rules (Article 2(9)), the Digital Services Act liability rules for intermediary services (Article 2(5)) or data protection law (Article 2(7)). Member States may keep or introduce rules that protect workers more favourably (Article 2(11)).

Scope is not the same as obligations

An in-scope organisation still has to work out which obligations apply. That depends on the risk tier.

TierWhat it means for an in-scope organisation
ProhibitedBanned under Article 5 since 2 February 2025. Two prohibitions added by the Digital Omnibus, on AI that generates non-consensual intimate imagery of identifiable people and on AI that generates child sexual abuse material, apply from 2 December 2026
High-riskFull Chapter III duties for providers and deployers, from 2 December 2027 (Annex III) or 2 August 2028 (Annex I)
Limited riskTransparency duties under Article 50, applying since 2 August 2026
Minimal riskNo specific obligations in the Act, apart from the AI literacy duty in Article 4 that applies to every provider and deployer

When the rules reach you: Systems already on the market get a transition rule. Under Article 111(2), a high-risk system placed on the market or put into service before the high-risk rules start applying is caught only if its design is significantly changed from that date. The exception is high-risk AI intended for public authorities, which has to comply by 2 August 2030 in any case. That is why “significant change” has become a planning question for existing models. For the full calendar, see the Compliance Timeline.

SMEs and small mid-caps

SMEs and small mid-cap companies are in scope. The relief concerns how obligations are met, not whether they apply. The Digital Omnibus extended several SME measures to small mid-caps (SMCs). The Commission’s 2025 recommendation defines an SMC as a company that is not an SME, has fewer than 750 staff and has annual turnover of up to €150 million or a balance sheet total of up to €129 million. For SMCs the Act now provides:

  • a simplified technical documentation form that notified bodies must accept (Article 11(1));
  • quality management proportionate to the size of the organisation (Article 17(2));
  • priority access to the EU-level regulatory sandbox that the AI Office may establish (Article 57(3a));
  • a duty on Member States to take SMC interests and economic viability into account when imposing penalties (Article 99(1));
  • a cap that limits each fine under Article 99(4) and (5) to the lower of the percentage and the fixed amount (Article 99(6a)).

Scope check

  • Every AI system in use or development is listed, with a note on what it does
  • Each system is tested against the seven elements of the AI system definition and the result is recorded
  • A role (provider, deployer, importer, distributor) is assigned for each system
  • The link to the EU is identified: placement on the market, establishment, or output used in the EU
  • Any exclusion relied on is checked against its condition, including “exclusively” and “sole purpose”
  • Systems already on the market before the high-risk start dates are marked, with the design changes that would count as significant
  • SME or small mid-cap status is checked for the documentation, quality management and penalty relief
  • Every in-scope system moves on to risk classification

Frequently asked questions

Not quite. Article 74’s documentation and source code powers are aimed primarily at providers, since they hold the technical documentation, datasets, and source code an authority needs to assess compliance. Deployers face a narrower cooperation duty under Article 26(12), and in practice are asked for whatever they specifically hold: the six-month logs under Article 26(6), oversight records under Article 26(2), and the Fundamental Rights Impact Assessment where one applies. A deployer that has substantially modified the system stops being treated as a deployer at all for that system, taking on the full provider-level exposure instead.

It’s best to have a structured compliance checklist in place: start with a complete inventory of every agent and model, classify each by risk, and attach model-level monitoring, explainability, human oversight, and automatic audit trails mapped to the Act’s requirements. A dedicated governance platform such as Deeploy centralises this so the evidence exists before an audit asks for it.

In a defined set of cases. Article 86 applies to decisions taken by a deployer on the basis of the output of an Annex III high-risk system (excluding point 2) that produce legal effects or similarly significantly affect the person in a way they consider adverse to their health, safety or fundamental rights. It applies only to the extent the right is not already provided for under other Union law.

Article 11 allows SMEs and start-ups to submit Annex IV information in a simplified form that the European Commission is required to publish, and notified bodies must accept it for conformity assessment. As of mid-2026, that simplified form has not yet been published. The content required is unchanged; only the format and level of detail will differ once it exists.

Post-market monitoring is the ongoing obligation to track how a high-risk AI system performs once it is deployed. Under Article 72, providers must establish a post-market monitoring plan before deployment covering the metrics to be tracked, the data to be collected, the review frequency, and the conditions that would trigger corrective action or re-assessment. Deployers must monitor the system’s operation and report relevant findings back to the provider.

Yes. If for example a chatbot or content-generation tool is redeployed into one of the Annex III high-risk domains, or substantially modified in a way that changes its risk profile, its classification has to be reassessed rather than assumed to stay limited risk indefinitely.

Disclaimer

This is general information, not legal advice. Please consult your legal or compliance team to confirm your organisation's specific obligations.

Reading about compliance
is step one. Operating
it is Deeploy

See how teams use Deeploy to monitor, document and govern their AI against the EU AI Act.