Under the EU AI Act the same AI system carries very different obligations depending on whether your organisation is its provider or its deployer, and many organisations that think they are only deployers have quietly become providers by rebranding, fine-tuning or repurposing a system. This guide explains how Regulation (EU) 2024/1689 defines the operator roles, the three Article 25 triggers that turn a deployer, distributor or importer into a provider, what each role must do for high-risk systems and for transparency, and how the Digital Omnibus on AI has moved the dates. It is written for security, risk, legal and AI platform teams who need to decide, system by system, which role they hold. Status checked on 27 October 2026 against the European Commission's AI Act Service Desk consolidated text (as at 27 July 2026) and the Commission's AI Act policy pages; deep coverage of the high-risk requirements themselves is in the EU AI Act High-Risk AI Systems Guide.
Key takeaways
- Role is decided per system, not per company. One organisation can be the provider of an internal tool it built, the deployer of a vendor's HR screening system and the provider of a vendor model it fine-tuned and released under its own name.
- Three acts make you a provider of a high-risk system. Putting your name or trademark on it, making a substantial modification, or changing the intended purpose of a system (including a general-purpose AI system) so that it becomes high-risk (Article 25(1)).
- Providers carry the conformity burden; deployers carry the operational one. Providers need a quality management system, documentation, conformity assessment, CE marking and registration (Article 16); deployers must follow instructions, assign competent human oversight, control input data, monitor, keep logs for at least six months and inform affected people (Article 26).
- The high-risk dates have moved. As amended by the Digital Omnibus on AI, Chapter III Sections 1 to 3 apply from 2 December 2027 for Annex III systems and 2 August 2028 for Annex I product systems (Article 113).
- Penalties apply to both roles. Breaches of provider obligations (Article 16) and deployer obligations (Article 26) are both in the band of up to EUR 15 million or 3% of worldwide annual turnover, whichever is higher (Article 99(4)).
The operator roles
Article 3 defines six operator roles; "operator" is the umbrella term covering a provider, product manufacturer, deployer, authorised representative, importer or distributor (Article 3(8)).
| Role | Article 3 definition (summary) | Typical example |
|---|---|---|
| Provider (3(3)) | Develops 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, whether paid or free | A software vendor selling a credit-scoring system; a bank that builds its own fraud model for internal use |
| Deployer (3(4)) | Uses an AI system under its authority, except in the course of a personal non-professional activity | An employer using a vendor's CV screening tool |
| Authorised representative (3(5)) | Located or established in the EU, with a written mandate from a provider to carry out its obligations | EU representative of a US model provider |
| Importer (3(6)) | Located or established in the EU and places on the market a system bearing the name or trademark of a person established outside the EU | EU reseller importing a non-EU vendor's system |
| Distributor (3(7)) | Any other person in the supply chain, other than provider or importer, that makes a system available on the EU market | A marketplace or systems integrator reselling a system unchanged |
Two further definitions drive role changes. "Putting into service" includes supply of a system for first use directly to the deployer or for the provider's own use in the EU (Article 3(11)), which is why an organisation that builds and uses its own system is its provider. "Substantial modification" is a change made after placing on the market or putting into service (Article 3(23)) that affects compliance or the intended purpose; the full definition sets the test.
How deployers become providers: Article 25
Article 25(1) states that any distributor, importer, deployer or other third party is considered a provider of a high-risk AI system, with the provider obligations of Article 16, in three circumstances.
| Trigger | Article | Common real-world case | Watch for |
|---|---|---|---|
| Name or trademark | 25(1)(a) | White-labelling a vendor's high-risk system as your own product | Applies "without prejudice to contractual arrangements stipulating that the obligations are otherwise allocated", so the contract matters |
| Substantial modification | 25(1)(b) | Retraining, changing model components or decision logic of a high-risk system already on the market, where it remains high-risk under Article 6 | Changes the provider pre-determined in its conformity assessment are not substantial; document which changes your team makes |
| Change of intended purpose | 25(1)(c) | Using a general-purpose chatbot or model to rank job applicants, assess creditworthiness or decide access to public services, so it becomes high-risk under Article 6 | The most common accidental route: a business unit repurposes a general tool without telling compliance |
- What happens to the original provider (25(2)): once another party becomes the provider, the initial provider is no longer considered the provider of that specific system, but must cooperate closely with the new provider and make the necessary information, technical access and assistance available. Article 25 has been amended by the Digital Omnibus on AI, so read the current consolidated paragraph 2 for the conditions that apply.
- Written agreements (25(4)): the provider of a high-risk system and a third party that supplies AI models, tools, services, components or processes used in it must specify by written agreement the information, capabilities, technical access and assistance needed for compliance. This does not apply to third parties making free and open-source components publicly available, other than general-purpose AI models.
- Product manufacturers (25(3)): where a high-risk system is a safety component of a product under the Union harmonisation legislation in Annex I Section A, the product manufacturer is the provider if the system is placed on the market or put into service under its name or trademark.
- Trade secrets (25(5)): the cooperation duties are without prejudice to protecting intellectual property, confidential business information and trade secrets.
Fine-tuning a general-purpose model
Fine-tuning raises two separate questions. First, is the resulting AI system high-risk because of its intended purpose? If so, you are its provider under Article 3(3) or Article 25(1)(c). Second, have you become the provider of a modified general-purpose AI model? The Commission's guidelines on the scope of GPAI provider obligations state that only those making significant modifications to a model need to comply with the GPAI provider obligations and that minor changes do not trigger them. GPAI provider obligations applied from 2 August 2025, Commission enforcement powers including fines apply from 2 August 2026, and models placed on the market before 2 August 2025 must comply by 2 August 2027. See the EU General-Purpose AI (GPAI) Code of Practice Guide.
Provider and deployer obligations side by side
| Area | Provider of a high-risk system | Deployer of a high-risk system |
|---|---|---|
| Core requirements | Ensure the system meets Chapter III Section 2 requirements (Art. 16(a)) | Use the system in accordance with the instructions for use, with appropriate technical and organisational measures (Art. 26(1)) |
| Governance | Quality management system under Art. 17 (16(c)); documentation under Art. 18 (16(d)) | Assign human oversight to people with the necessary competence, training, authority and support (26(2)) |
| Data | Data and data governance requirements in Section 2 | Where the deployer controls input data, ensure it is relevant and sufficiently representative for the intended purpose (26(4)) |
| Logs | Keep automatically generated logs under Art. 19 when under its control (16(e)) | Keep logs under its control for a period appropriate to the purpose, at least six months, unless other law provides otherwise (26(6)) |
| Market entry | Conformity assessment (Art. 43), EU declaration of conformity (Art. 47), CE marking (Art. 48), registration (Art. 49(1)) (16(f) to (i)) | Public authority deployers must register their use and must not use a system that is not registered in the EU database (26(8)) |
| Monitoring and incidents | Corrective actions and information duties (Art. 20, via 16(j)); post-market monitoring and serious incident reporting | Monitor operation; if the system presents a risk, inform the provider or distributor and the market surveillance authority and suspend use; report serious incidents to the provider first (26(5)) |
| People affected | Name and contact on the system or documentation (16(b)); accessibility requirements (16(l)) | Inform workers' representatives and affected workers before workplace use (26(7)); inform people subject to Annex III decisions that the system is used (26(11)) |
| Impact assessment | Provide the information deployers need under Art. 13 | Use Art. 13 information for the GDPR Article 35 DPIA (26(9)); certain deployers must carry out a fundamental rights impact assessment (Art. 27) |
| Authorities | Demonstrate conformity on a reasoned request (16(k)) | Cooperate with competent authorities (26(12)) |
Article 26(5) and 26(6) contain specific provisions for financial institutions, which can meet the monitoring and log-keeping duties through their existing financial services governance and documentation rules.
Fundamental rights impact assessment
Before deploying a high-risk system referred to in Article 6(2), bodies governed by public law, private entities providing public services and deployers of the Annex III point 5(b) and (c) systems (creditworthiness and credit scoring, and life and health insurance risk assessment and pricing) must assess the impact on fundamental rights (Article 27(1)); systems in Annex III point 2 (critical infrastructure) are excluded. The assessment covers the deployer's processes, period and frequency of use, affected groups, specific risks of harm, human oversight measures and the response if risks materialise, and the results are notified to the market surveillance authority. As amended, Article 27 lets the deployer cross-reference or include relevant parts of an existing DPIA.
Obligations that apply whatever the risk class
| Obligation | Provider | Deployer | Applies from |
|---|---|---|---|
| AI literacy (Art. 4, as amended) | Take measures to support the AI literacy of staff and others acting on its behalf | Same | 2 February 2025 (Chapter I) |
| Prohibited practices (Art. 5) | Must not place on the market | Must not use | 2 February 2025; new Article 5 points added by the Digital Omnibus from 2 December 2026 |
| Interaction disclosure (Art. 50(1)) | Design systems that interact with people so they know they are dealing with AI, unless obvious | No | 2 August 2026 |
| Machine-readable marking of synthetic content (Art. 50(2)) | Mark generated audio, image, video and text as artificially generated | No | 2 August 2026; transitional deadline of 2 December 2026 for certain systems placed on the market before 2 August 2026 |
| Emotion recognition and biometric categorisation (Art. 50(3)) | No | Inform the people exposed | 2 August 2026 |
| Deepfakes and public-interest text (Art. 50(4)) | No | Disclose that content is artificially generated or manipulated, with limited exceptions | 2 August 2026 |
The AI literacy duty is covered in the EU AI Act AI Literacy Obligations (Article 4) Guide and prohibited practices in the EU AI Act Prohibited Practices Guide.
Current status and dates after the Digital Omnibus
The AI Act entered into force on 1 August 2024. The Commission adopted its Digital Omnibus on AI proposal on 19 November 2025, and a political agreement was reached on 7 May 2026. The AI Act Service Desk consolidated text as at 27 July 2026 marks the affected provisions as amended by the Digital Omnibus on AI. Dates below are from Article 113 as shown in that consolidated text; check the Official Journal for the amending act before relying on them in a contract or compliance plan.
| Date | What applies |
|---|---|
| 2 February 2025 | Chapters I and II: definitions, AI literacy, prohibited practices |
| 2 August 2025 | Notified bodies (Chapter III Section 4), general-purpose AI models (Chapter V), governance (Chapter VII), penalties (Chapter XII) and Article 78 |
| 27 July 2026 | Articles 102 to 110 (new point added by the Digital Omnibus) |
| 2 August 2026 | General date of application, including Article 50 transparency |
| 2 December 2026 | New Article 5 prohibitions added by the Digital Omnibus |
| 2 December 2027 | High-risk requirements (Chapter III Sections 1 to 3) for systems under Article 6(2) and Annex III |
| 2 August 2028 | High-risk requirements for systems under Article 6(1) and Annex I (AI in regulated products) |
Penalties by role
- Prohibited practices (Art. 99(3)): up to EUR 35 million or 7% of total worldwide annual turnover for the preceding financial year, whichever is higher.
- Operator obligations (Art. 99(4)): up to EUR 15 million or 3%, whichever is higher, covering providers (Art. 16), authorised representatives (Art. 22), importers (Art. 23), distributors (Art. 24), Article 25(2) and (4), deployers (Art. 26), notified bodies and transparency (Art. 50).
- Incorrect, incomplete or misleading information (Art. 99(5)): up to EUR 7.5 million or 1%, whichever is higher.
- SMEs and start-ups (Art. 99(6)): the lower of the two amounts applies; the amended text adds a similar rule for small mid-cap companies for paragraphs 4 and 5.
Where identity fits
Deployer duties are partly identity duties. Human oversight under Article 26(2) means named, trained people with authority to intervene, which needs role assignments, access rights and an audit trail of who overrode what. Log retention under Article 26(6) needs the system's actions tied to identities: when an AI agent acts through a service account or OAuth grant, the log must show which agent, under whose authority and with what scope. Treat every deployed AI system's credentials as non-human identities with an owner. See the AI Agent Audit Trails for Regulators Guide and the Agentic AI Compliance Guide.
How to determine and evidence your role
- Inventory AI systems: list every system built, bought, embedded in SaaS or accessed by API, with its business use. See the EU AI Act Compliance Guide.
- Classify each use: prohibited, high-risk (Annex I or Annex III), transparency-only or minimal, based on the actual use, not the vendor's marketing.
- Assign the role per system: record whether you developed it, have it developed, sell or use it under your name, or only use it.
- Test the Article 25 triggers: for every vendor system, ask whether you rebranded it, modified it substantially or changed its intended purpose. Record the reasoning and who approved it.
- Control change: require compliance sign-off before retraining, fine-tuning or repurposing any system, because each can move you into the provider role.
- Contract: use Article 25(4) written agreements and procurement terms to obtain the instructions for use, Article 13 information, log access and incident cooperation you need. See the AI Procurement Checklist.
- Operate deployer controls: oversight roles, log retention of at least six months, input data checks, worker information, notices to affected people, FRIA where required.
- Review annually and on change: re-run the role test when a system's use, vendor or model changes.
Decision guidance
- Rebrand or not: if you want to sell a vendor's high-risk system under your own brand, either accept the provider role and its conformity work, or negotiate a contractual allocation and keep the vendor's name visible.
- Build or buy: building an internal high-risk system makes you its provider even if it is never sold; buying keeps you a deployer as long as you use it as instructed.
- Small organisations: most will be deployers; the realistic risk is accidental provider status through repurposing a general-purpose tool for an Annex III use.
Detection and measurement
- Percentage of AI systems with a recorded role decision and Article 25 assessment.
- Number of retraining, fine-tuning or repurposing changes made without compliance sign-off.
- High-risk deployments with named, trained oversight owners and log retention of at least six months verified.
- Vendor contracts for high-risk systems containing Article 13 information, log access and incident cooperation terms.
- FRIAs completed and notified where Article 27 applies.
Framework mapping
| Framework | Reference | Link to provider and deployer duties |
|---|---|---|
| ISO/IEC 42001:2023 | AI management system | Supports a provider's quality management system and a deployer's governance of use |
| NIST AI RMF 1.0 | Govern, Map, Measure, Manage functions | Role assignment and accountability (Govern); intended purpose and context (Map). See the NIST AI RMF Guide |
| NIST CSF 2.0 | GV.RR-02 roles and responsibilities; GV.SC-05 supplier requirements in contracts | Role decisions; Article 25(4) agreements |
| GDPR | Article 35 DPIA | Article 26(9) and Article 27 cross-reference |
Common failure patterns
- Assuming "we only use it": a team repurposes a general-purpose assistant to screen candidates, which is a change of intended purpose under Article 25(1)(c).
- White-labelling without a contract: the product carries your brand but the vendor holds the technical documentation you would need as provider.
- Uncontrolled fine-tuning: data science teams retrain vendor systems as routine maintenance with no record of whether the change is substantial.
- Oversight on paper only: a named overseer with no training, no authority to stop the system and no access to its outputs.
- Logs deleted early: default SaaS retention shorter than six months, or logs held only by the vendor with no contractual access.
- Planning to the old dates: programmes still built around 2 August 2026 for Annex III high-risk systems, or, the reverse, treating the moved dates as a reason to stop preparation.
Practitioner checklist
- Every AI system has a recorded role (provider, deployer, importer, distributor) and the reasoning.
- Each vendor system has an Article 25(1) assessment: name or trademark, substantial modification, change of intended purpose.
- Retraining, fine-tuning and repurposing require compliance sign-off.
- Contracts for high-risk systems cover instructions for use, Article 13 information, log access and incident cooperation.
- High-risk deployments have trained oversight owners with authority to intervene.
- Logs under your control are kept for at least six months and tied to identities.
- Workers, affected people and, where required, the market surveillance authority (FRIA) are informed.
- Transparency duties under Article 50 are mapped to the right role for each system.
- Compliance plans use the Article 113 dates as amended, with a dated note of the last status check.
Standards and references
- EUR-Lex: Regulation (EU) 2024/1689 (Artificial Intelligence Act)
- European Commission AI Act Service Desk: Article 25 Responsibilities along the AI value chain
- European Commission AI Act Service Desk: Article 26 Obligations of deployers of high-risk AI systems
- European Commission AI Act Service Desk: Article 16 Obligations of providers of high-risk AI systems
- European Commission AI Act Service Desk: Article 113 Entry into force and application
- European Commission: AI Act regulatory framework
- European Commission: Guidelines for providers of general-purpose AI models
- European Commission AI Act Service Desk: Article 99 Penalties
Related NHI Mgmt Group resources: EU AI Act High-Risk AI Systems Guide · EU AI Act Compliance Guide · Third-Party AI Risk Guide · Agentic AI Compliance Guide. This guide summarises the provider and deployer obligations of Regulation (EU) 2024/1689 as amended and shown in the Commission's consolidated text on 27 October 2026, and is not legal advice.