Guide
The EU AI Act for companies: what applies when
A practical reading of Regulation (EU) 2024/1689 as amended by the 2026 Digital Omnibus on AI: which role you have for each system, which risk class it falls into, what that requires, and from which date.
The short version
The AI Act lays down rules for AI systems placed on the market, put into service or used in the EU. It was published in the Official Journal on 12 July 2024, entered into force on 1 August 2024 and applies in stages (Art. 113). Regulation (EU) 2026/1744, known as the Digital Omnibus on AI, amended it: published on 24 July 2026, in force since 27 July 2026. Among other things it moved the dates for high-risk systems back.
What you have to do depends on two questions, asked per AI system: what is your role for the system, and which risk class does it fall into. This guide follows that order. Article numbers refer to the consolidated text.
What applies when
| Date | What applies | Provision |
|---|---|---|
| 2 Feb 2025 | General provisions, including AI literacy (Art. 4), and the prohibited practices (Art. 5) | Art. 113(a) |
| 2 Aug 2025 | Notified bodies (Chapter III Section 4), obligations for providers of general-purpose AI models (Chapter V), governance (Chapter VII), penalties (Chapter XII, except Art. 101) and confidentiality (Art. 78) | Art. 113(b) |
| 27 Jul 2026 | The Digital Omnibus enters into force; Arts. 102 to 110, which amend other Union acts, apply | Art. 113(d) |
| 2 Aug 2026 | The general date of application: everything not listed elsewhere, including the transparency obligations of Art. 50 and the fines for providers of general-purpose AI models (Art. 101) | Art. 113 |
| 2 Dec 2026 | Two new prohibitions in Art. 5(1)(ba) and (bb); machine-readable marking under Art. 50(2) for generative systems placed on the market before 2 Aug 2026 | Art. 113(a), Art. 111(4) |
| 2 Aug 2027 | General-purpose AI models placed on the market before 2 Aug 2025 must comply; national AI regulatory sandboxes must be operational | Art. 111(3), Art. 57(1) |
| 2 Dec 2027 | Requirements and obligations for high-risk systems in the Annex III areas | Art. 113(c)(i) |
| 2 Aug 2028 | The same for high-risk systems under the product legislation in Annex I | Art. 113(c)(ii) |
| 2 Aug 2030 | High-risk systems intended for use by public authorities that were already on the market | Art. 111(2) |
Other high-risk systems that were already on the market before their date of application fall under the Act only once their design changes significantly (Art. 111(2)).
Your role decides your obligations
Most obligations attach to a role, not to a company as a whole. One company often has different roles for different systems, so record the role per system.
- Provider (Art. 3(3)): develops an AI system or a 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 for payment or free of charge. Putting into service includes supplying a system “for own use” (Art. 3(11)), so a company that builds a system and uses it internally is its provider.
- Deployer (Art. 3(4)): uses an AI system under its authority, except in a personal, non-professional activity. A company that buys or licenses a system and uses it at work is a deployer.
- Importer (Art. 3(6)) and distributor (Art. 3(7)): bring a system from outside the EU onto the market, or make it available further down the supply chain.
- Operator (Art. 3(8)): the umbrella term for all of these and a few more.
Roles can change. Under Art. 25(1), a distributor, importer, deployer or other third party becomes the provider of a high-risk system if it puts its name or trademark on the system, makes a substantial modification, or changes the intended purpose so that the system becomes high-risk.
Risk classes
Prohibited practices (Art. 5)
Banned since 2 February 2025: manipulative or deceptive techniques that cause significant harm; exploiting vulnerabilities due to age, disability or a social or economic situation; social scoring; predicting crimes based solely on profiling or personality traits; untargeted scraping of facial images for recognition databases; emotion recognition at the workplace and in education, except for medical or safety reasons; biometric categorisation to infer sensitive characteristics such as political opinions or sexual orientation; and, with narrow exceptions, real-time remote biometric identification in public spaces for law enforcement.
From 2 December 2026 two more apply: AI systems that generate intimate or sexual images of an identifiable person without that person's explicit consent, and systems that generate child sexual abuse material (Art. 5(1)(ba) and (bb)).
High-risk AI systems (Art. 6)
There are two routes into the high-risk class:
- Products (Art. 6(1), Annex I): the AI system is a product, or a safety component of a product, covered by the EU product legislation in Annex I, such as medical devices or machinery, and that product must undergo a third-party conformity assessment.
- Use cases (Art. 6(2), Annex III): the system is used in one of eight areas: biometrics; critical infrastructure; education and vocational training; employment, workers' management and access to self-employment; access to essential private and public services and benefits, such as creditworthiness and life and health insurance pricing; law enforcement; migration, asylum and border control; administration of justice and democratic processes.
An Annex III system is not high-risk if it does not pose a significant risk of harm, for example because it only performs a narrow procedural task or a preparatory task (Art. 6(3)). A system that profiles natural persons is always high-risk. A provider that relies on this exception documents its assessment and still registers the system (Art. 6(4), Art. 49(2)).
Transparency obligations (Art. 50)
- Providers design systems that interact with people so those people know they are dealing with AI, unless that is obvious (Art. 50(1)).
- Providers of systems that generate audio, images, video or text mark the output in a machine-readable, detectable way (Art. 50(2)).
- Deployers of emotion recognition or biometric categorisation inform the people exposed to it (Art. 50(3)).
- Deployers disclose deep fakes, and AI-generated text published to inform the public on matters of public interest unless a person with editorial responsibility reviewed it (Art. 50(4)).
Minimal risk
Everything else, which covers most AI in use. There are no specific requirements beyond the AI literacy measures of Art. 4; voluntary codes of conduct are possible (Art. 95).
General-purpose AI models (Arts. 51 to 55)
A separate track for those who place general-purpose models on the market: technical documentation, information for downstream providers, a copyright policy and a summary of training content (Art. 53), and more for models with systemic risk (Art. 55). A company that builds an application on such a model is usually not the model's provider, but it can be the provider of the AI system it builds.
What high-risk systems require
Providers
- a risk management system (Art. 9) and data governance for training, validation and test data (Art. 10);
- technical documentation along Annex IV (Art. 11); SMEs, including start-ups, and small mid-cap enterprises may provide it in a simplified form (Art. 11(1));
- automatic logging of events over the system's lifetime (Art. 12);
- instructions and information for deployers (Art. 13) and measures for human oversight (Art. 14);
- appropriate accuracy, robustness and cybersecurity (Art. 15);
- a quality management system (Art. 17), which SMEs may implement in a simplified manner for certain elements (Art. 63(1));
- conformity assessment (Art. 43), registration (Art. 49), post-market monitoring (Art. 72);
- reporting serious incidents to the market surveillance authority (Art. 73): no later than 15 days after becoming aware, 10 days if a person died, and 2 days for a widespread infringement or a serious disruption of critical infrastructure.
Deployers (Art. 26)
- use the system according to its instructions (Art. 26(1)), and give human oversight to people with the competence, training and authority for it (Art. 26(2));
- where you control the input data, make sure it is relevant and sufficiently representative (Art. 26(4));
- monitor the operation, inform the provider of risks and serious incidents, and suspend use where needed (Art. 26(5));
- keep the logs the system generates, as far as they are under your control, for at least six months (Art. 26(6));
- inform workers' representatives and affected workers before using a high-risk system at the workplace (Art. 26(7));
- inform people that a high-risk system from Annex III makes or assists decisions about them (Art. 26(11)).
A fundamental rights impact assessment (Art. 27) is required from deployers that are bodies governed by public law or private entities providing public services, and from deployers of systems for creditworthiness or credit scoring and for risk assessment and pricing in life and health insurance (Annex III, points 5(b) and 5(c)).
Fines
| Infringement | Up to |
|---|---|
| Prohibited practices (Art. 99(3)) | €35 million or 7% of total worldwide annual turnover |
| Most other obligations of operators and notified bodies (Art. 99(4)) | €15 million or 3% |
| Incorrect, incomplete or misleading information to notified bodies or authorities (Art. 99(5)) | €7.5 million or 1% |
For SMEs, including start-ups, the lower of the two amounts applies (Art. 99(6)). Since the Digital Omnibus, the same holds for small mid-cap enterprises for the fines in Art. 99(4) and (5) (Art. 99(6a)). Fines on providers of general-purpose AI models are imposed by the Commission (Art. 101).
What the 2026 Digital Omnibus changed
The Commission proposed the Digital Omnibus on AI on 19 November 2025. Parliament approved the agreed text on 16 June 2026, the Council adopted it on 29 June 2026, and it was published as Regulation (EU) 2026/1744 on 24 July 2026. The changes most companies will notice:
- Later high-risk dates. Annex III systems from 2 December 2027 instead of 2 August 2026; Annex I products from 2 August 2028 instead of 2 August 2027.
- AI literacy reworded. Providers and deployers take measures to support the AI literacy of their staff; the article now says it does not require them to guarantee a specific level. The Commission and Member States support companies, in particular SMEs (Art. 4).
- Two new prohibitions from 2 December 2026, on non-consensual intimate images and child sexual abuse material.
- More time for marking AI-generated content for generative systems already on the market before 2 August 2026: until 2 December 2026 (Art. 111(4)).
- Relief for smaller companies. A new category of small mid-cap enterprises, simplified technical documentation (Art. 11(1)), a simplified quality management system for SMEs (Art. 63(1)) and lower fine caps (Art. 99(6) and (6a)).
- Bias detection. A new Art. 4a on processing special categories of personal data to detect and correct bias.
- Sandboxes. National AI regulatory sandboxes by 2 August 2027 (Art. 57(1)), and an AI Office sandbox at Union level.
- Existing high-risk systems. Covered only after significant design changes, except systems for public authorities, which must comply by 2 August 2030 (Art. 111(2)).
Providers that use the Art. 6(3) exception still register their systems under Art. 49(2), with fewer data fields than before.
What to do now
As of November 2026, the prohibitions, AI literacy and the transparency obligations already apply. The high-risk requirements follow in December 2027 and August 2028, and the technical work behind them takes longer than a year.
List every AI system
Including the ones you buy and the ones built on general-purpose models, with purpose, owner and the people affected.
Record your role per system
Provider, deployer, importer or distributor, and check Art. 25 for systems you rebrand, modify substantially or repurpose.
Rule out prohibited practices
Art. 5 has applied since February 2025, and two more prohibitions follow on 2 December 2026.
Document AI literacy measures
Training and guidance that fit each group's role and the systems it works with (Art. 4).
Classify each system
Check Annex I and Annex III, document every Art. 6(3) assessment, and register those systems.
Meet the transparency duties
Disclosure for chatbots, marking of generated content, and labelling of deep fakes (Art. 50).
Start the high-risk work early
Logging with at least six months' retention, data governance, testing, human oversight, technical documentation and an incident process with 15, 10 and 2 day deadlines. Deployers add worker information and, where required, a fundamental rights impact assessment.
Keep the evidence as you go
Records written while you build and run a system are easier to defend than documents reconstructed before an audit.
Where EUVYRA fits
EUVYRA keeps the inventory of AI systems, including bought ones, with the role per system. Its classification questionnaire follows Arts. 5, 6 and 50 and Annexes I and III, with a second person approving the result. Controls mapped to Arts. 4, 5, 9 to 15, 17, 19, 26, 27, 50, 53, 72 and 73 check themselves against your runs, evaluations, releases and logs, and the incident clocks follow Art. 73 alongside NIS2, DORA and the GDPR. Governance in EUVYRA
EUVYRA collects evidence and supports review. It does not certify conformity, and its classification is a documentation aid, not legal advice.
Frequent questions
We only use AI tools from vendors. Does the AI Act apply to us?
Yes, as a deployer. AI literacy measures (Art. 4) apply already. Deployers of emotion recognition, biometric categorisation or deep fakes have transparency duties (Art. 50(3) and (4)), and if you use a high-risk system, such as software that screens job applications, the deployer obligations of Art. 26 apply from 2 December 2027.
We built a chatbot on a general-purpose model. Are we a provider?
Of the chatbot, usually yes: you put an AI system into service under your own name, and that includes own use (Art. 3(3) and 3(11)). People must be able to tell they are talking to AI unless it is obvious (Art. 50(1)). The obligations for the underlying model (Art. 53) sit with whoever places that model on the market.
When do the high-risk obligations start?
After the Digital Omnibus, on 2 December 2027 for systems in the Annex III areas and on 2 August 2028 for systems covered by the product legislation in Annex I (Art. 113(c)).
Is AI for recruitment high-risk?
Systems for recruiting or selecting people, such as placing targeted job ads, filtering applications or evaluating candidates, are listed in Annex III, point 4, unless the Art. 6(3) exception applies. Profiling people always keeps a system high-risk.
Does the AI Act replace the GDPR?
No, both apply side by side. Deployers use the information a provider gives under Art. 13 for their data protection impact assessment (Art. 26(9)), and the Digital Omnibus added Art. 4a on processing special categories of data for bias detection.
How long must deployers keep logs?
At least six months, as far as the logs are under their control, unless other EU or national law says otherwise (Art. 26(6)). Financial institutions keep them as part of their financial-sector documentation.
Sources
Checked on 13 November 2026. Article numbers and dates follow the consolidated text of Regulation (EU) 2024/1689 as amended by Regulation (EU) 2026/1744.
- Regulation (EU) 2024/1689 (Artificial Intelligence Act), OJ L, 2024/1689, 12.7.2024
- Regulation (EU) 2026/1744 (Digital Omnibus on AI), OJ L, 2026/1744, 24.7.2026
- European Commission, AI Act Service Desk: consolidated text as at 27 July 2026, Article 113
- Consolidated text, Article 3 (definitions)
- Consolidated text, Article 26 (obligations of deployers of high-risk AI systems)
- Consolidated text, Article 50 (transparency obligations)
- Consolidated text, Annex III (high-risk areas)
- European Parliament, Legislative Observatory: procedure 2025/0359(COD)