Questions and answers

A provider develops an AI system or general-purpose AI model, or has one developed, and places it on the market or puts the system into service under its own name or trademark. Development, commissioning, branding and first supply or use all matter. Writing the code alone does not decide the role.

A deployer uses an AI system under its authority, except for purely personal, non-professional use. A company using a supplier's system can be the deployer while the supplier remains the provider. Check who controls the use, operating process, users, relevant input data and oversight.

Operator is the umbrella term for a provider, product manufacturer, deployer, authorised representative, importer or distributor. Identify the specific role before assigning duties.

An authorised representative is EU-based and has a written provider mandate. An importer first places on the EU market a system bearing a third-country person's name or trademark. A distributor makes a system available on the EU market and is neither its provider nor importer.

A product manufacturer is treated as the provider of a high-risk AI system that is a safety component of a product covered by the Union harmonisation legislation in Section A of Annex I when the system is placed on the market together with the product under the manufacturer's name or trademark, or is put into service under that name or trademark after the product has been placed on the market.

Article 25 has three triggers for a system already on the market or in service. A party becomes the provider if it puts its name or trademark on a high-risk system, substantially modifies a high-risk system while it remains high-risk, or changes a non-high-risk system's intended purpose so it becomes high-risk. The last trigger also covers general-purpose AI systems. The branding trigger preserves contractual arrangements allocating obligations differently.

The initial provider then ceases to be the provider of that specific system for AI Act purposes. It must cooperate closely with the new provider and supply necessary information, technical access and reasonably expected assistance. That cooperation duty does not apply where the initial provider clearly said its system must not be changed into a high-risk system.

Article 25, as amended by Regulation (EU) 2026/1744, specifies that this cooperation includes, where relevant, technical documentation sufficient to assess Article 16 compliance, known limitations and failure modes, and targeted technical access for testing and validation.

Use contracts to agree work, access, assistance, confidentiality and liability. Under amended Article 25(4), a high-risk system provider and a supplier of an integrated system, model, tool, service, component or process must specify the necessary information, capabilities, technical access and assistance in writing. Publicly accessible free and open-source tools, services, processes and components are excluded from that paragraph; general-purpose AI models are not.

Keep the model provider separate from the provider of a system built with it. A general-purpose AI model provider does not automatically provide every downstream system. Annex XII requires model information for system builders.

Commission guidance gives three examples: developing and marketing a general-purpose model; commissioning one and marketing it; or integrating a model into a system. The first two can make the organisation the model provider. The third can make it the system provider.

Modifying a model does not automatically make you its provider. Commission guidance looks for significant changes to generality, capabilities or systemic risk and gives an indicative training-compute criterion. Keep the change evidence and apply the guidance to the actual modification.

Keep a role matrix for each system and model version. Record the legal role, the facts that create it and the responsibility assigned. A single AI-owner field cannot describe a supplier relationship with several model and system roles.

For deployer duties, preserve records showing use according to instructions, human oversight assignment, monitoring based on instructions for use, incident escalation to provider/importer/distributor and authorities where relevant, logs under deployer control, worker information where workplace use is involved, and any required fundamental-rights impact assessment for covered high-risk deployments.

For Annex III point 1 biometric systems, a provider may choose Annex VI internal control or Annex VII notified-body assessment when it demonstrates compliance using the relevant harmonised standards or, where applicable, common specifications. Articles 40 and 41 set the standards and specifications rules.

Annex III point 1 is the biometrics category. It covers specified remote biometric identification systems, biometric categorisation systems using sensitive or protected characteristics or inferred attributes, and emotion recognition systems, subject to the precise Annex III wording and exclusions. It should not be confused with every system that processes a face, voice, or other biometric data.

For those biometric systems, Annex VII becomes mandatory if relevant standards and specifications are unavailable, a relevant harmonised standard is unused or only partly applied, or available common specifications are unused. It also applies to the restricted part where a harmonised standard was published with a restriction.

Annex III points 2-8 use Annex VI internal control. Article 43 requires no notified-body involvement for that route.

For high-risk AI systems covered by Union harmonisation legislation in Section A of Annex I, Article 43(3), as replaced by Regulation (EU) 2026/1744 from 27 July 2026, keeps the conformity route under the applicable product law and makes the AI Act Section 2 requirements part of that assessment. If that product law permits assessment without a third party when all relevant harmonised standards are applied, inclusion of a high-risk AI safety component does not by itself force third-party assessment. If the same system is also listed in Annex III, the product-law route controls.

For Article 6(1) systems related to Annex I Section B products, amended Article 2(2) limits which AI Act provisions apply directly. Use the relevant sector law instead of assuming the Annex VI/VII route applies. For Section A systems, amended Article 43(3) also includes Article 17 quality-management assessment in the product procedure.

Under Annex VI, the provider checks its own evidence. It must verify that the quality management system meets Article 17, examine the technical documentation against Chapter III Section 2, and check that design, development and post-market monitoring match that documentation.

Prepare the technical documentation before market placement or first use and keep it current. Article 11 requires enough information for authorities and notified bodies to assess compliance. Annex IV lists the contents: purpose, versions, development, data, oversight, testing, cybersecurity, risk management, standards, declaration and monitoring.

Under Annex VII, a notified body reviews both the provider's quality management system and the system's technical documentation. The application identifies the provider and all AI systems covered by that management system. Include technical documentation for each system, Article 17 management-system records, procedures to keep it effective, and a declaration that the same application was not filed with another body.

The body may ask for evidence or tests and carry out tests itself. If other reasonable means are insufficient, it may seek access to training and trained models, subject to intellectual-property and trade-secret protection. It issues a Union technical documentation assessment certificate if the system conforms, or refuses with reasons.

Approval brings continuing surveillance, including periodic audits of the quality management system. Notify the body of intended changes to that system or the covered-system list, and of AI-system changes that could affect compliance or intended purpose.

Prepare an EU declaration of conformity for each high-risk system under Article 47. It may be written, machine-readable, physical or electronically signed. Identify the system, state compliance with Chapter III Section 2, include Annex V information and translate it for relevant authorities. Keep it current and available to national competent authorities for 10 years.

Article 48 requires CE marking for high-risk AI systems. For digital high-risk AI systems, a digital CE marking can be used only when it is easily accessible through the system interface or through an accessible machine-readable code or other electronic means. Where a notified body was responsible for the Article 43 conformity assessment, its identification number follows the CE marking and is also indicated in promotional material that mentions CE conformity.

Article 49 registration is separate from CE marking and must be checked before release. Providers or authorised representatives register themselves and Annex III high-risk AI systems in the EU database before placing them on the market or putting them into service, except for point 2 of Annex III, which is registered at national level. Providers also register systems they concluded are not high risk under Article 6(3). Public authorities, Union institutions, bodies, offices, agencies, and persons acting on their behalf have deployer registration duties for covered Annex III systems before use.

Article 46 lets a market surveillance authority authorise a specific high-risk system for a limited period before assessment is complete. The reasons must meet its exceptional public-security, life-and-health, environmental or key-infrastructure conditions. Complete assessment without undue delay. The authority must grant the permission. For Annex I Section A products, Article 46(7) limits the exception to the applicable product-law route.

An AI system is machine-based, operates with some autonomy and infers from inputs how to generate outputs. Those outputs can include predictions, content, recommendations or decisions that influence a physical or virtual environment. Check all parts of Article 3(1), including explicit or implicit objectives and the possibility of adaptation after deployment.

A tool is less likely to be an AI system when it only executes rules defined solely by people, such as fixed validation checks, deterministic routing tables, hard-coded eligibility rules, or ordinary calculations with no learning, reasoning, modelling, or inference beyond basic data processing.

Use Article 3(1) and Recital 12 to classify the actual system. The Commission's definition guidelines explain its interpretation; they do not replace the legal definition. Keep evidence of how the tool infers its outputs. A supplier label such as AI-powered or automated does not settle the question.

Keep separate records for the general-purpose AI model (GPAI model) and the AI system built with it. The model supplies reusable capabilities. The system uses those capabilities for an intended purpose. A company that integrates the model into its own system can have system-provider duties for that integration.

Embedded and product-linked cases need two checks. First, decide whether the AI component itself is an AI system. Second, decide whether Article 6 makes it high-risk because it is a safety component, is itself a covered product, or falls into an Annex III use case.

Apply the amended safety-component tests in Regulation (EU) 2026/1744. A safety function has the intended purpose of preventing or mitigating health and safety risks. Solely non-safety user assistance, optimisation, efficiency, automation, convenience, or quality control does not qualify, but a failure or malfunction that would endanger health and safety still does. A product assessment required solely for non-health-and-safety risks, such as spectrum distribution or electromagnetic interference without a health-and-safety effect, does not satisfy Article 6(1)(b).

The Act can cover providers outside the EU when they place AI systems or general-purpose AI models on the EU market. It can also cover non-EU providers or deployers whose system outputs are used in the EU. Article 2 separately addresses EU deployers, importers, distributors, product manufacturers, authorised representatives and affected people in the EU.

Role classification is fact-specific and can change after launch. A distributor, importer, deployer, or other third party can become the provider of a high-risk AI system if it puts its name or trademark on the system, substantially modifies it, or changes the intended purpose so that the system becomes high-risk.

Scope exclusions also matter. The Regulation contains specific exclusions for personal non-professional use, certain scientific research and development activity, pre-market research, testing or development that respects applicable Union law, and some free and open-source releases. Those exclusions have conditions and do not create a general exemption for commercial deployment, prohibited practices, high-risk systems, or Article 50 uses. The Article 2(8) pre-market exclusion does not cover testing in real-world conditions.

Keep enough evidence for someone else to repeat the classification. Identify the software, model or system assessed; why it meets or falls outside the AI definition; its intended purpose; the EU connection; each actor's role; and the high-risk checks completed.

For high-risk and near-high-risk cases, align the evidence with Annex IV-style system description fields even if the team is still at classification stage: interactions with other hardware or software, form of supply, product integration, development methods, third-party components, output quality, monitoring, limits, and lifecycle changes.

Providers must establish and document monitoring proportionate to the system's technology and risks. Actively and systematically collect, record and analyse relevant performance data throughout its lifetime.

Keep the post-market monitoring plan in the Annex IV technical documentation. The 2026 amendment replaces the earlier implementing-act mechanism with Commission guidance and a template, due by 2 September 2027. It was published on 24 July 2026 and entered into force on 27 July. Until the guidance is available, you can use your own or a vendor's form to organise evidence; identify it as that form.

For high-risk systems already covered by Section A of Annex I product legislation, Article 72 allows the provider to integrate equivalent AI Act elements into an existing sectoral post-market system and plan. The same integration option applies to Annex III point 5 systems placed on the market or put into service by financial institutions subject to Union financial-services governance rules.

Deployers monitor use according to the provider's instructions and, where relevant, inform providers under Article 72. They supply important evidence even though the provider owns the plan.

If instructed use may create an Article 79(1) risk to health, safety or fundamental rights, the deployer must suspend use and inform the provider or distributor and relevant market surveillance authority without undue delay. For a serious incident, immediately inform the provider first, then the importer or distributor and relevant authorities. If the provider cannot be reached, apply Article 73 with the necessary role adjustments. The notification duty excludes sensitive operational data of law-enforcement deployers.

Give ordinary feedback, nonconformity, risks and serious incidents separate handling routes. Under Article 20, a provider that considers or has reason to consider a high-risk system nonconforming must immediately bring it into conformity, withdraw it, disable it or recall it, as appropriate.