EU AI Act obligations for companies that operate an AI agent
Contents
EU AI Act obligations for a company that operates an AI agent start with its role: under Article 3(4) of Regulation (EU) 2024/1689, EUR-Lex, whoever uses an AI system under its own authority is the deployer. The obligations attach to that role, graded by risk and with their own dates. Five steps put them in order:
- Settle the role: deployer, provider or both (Article 3(3) and (4), Article 25).
- Support AI literacy: Article 4, since 2 February 2025.
- Check transparency: Article 50, since 2 August 2026, for certain systems.
- Settle the classification: Article 6 and Annex III, for each Vorgang (one intake, one task, one destination) separately.
- Prepare the deployer obligations: Article 26, for Annex III high-risk systems from 2 December 2027.
The dates follow from Article 113 as amended by the AI Omnibus (consolidated text, EUR-Lex).
Obligations follow the role, and the role follows from Article 3
For day-to-day purposes the Act distinguishes two roles. The provider develops an AI system, or has it developed, and places it on the market or puts it into service under its own name or trademark (Article 3(3)). The deployer uses it under its own authority, except in a personal non-professional activity (Article 3(4)). One company can hold both roles at once.
The role can change. Under Article 25(1), EUR-Lex, a deployer is considered the provider of a high-risk system if it puts its name or trademark on it, makes a substantial modification to it, or changes a system's intended purpose so that it becomes high-risk. An example: a company that starts using an agent built for delivery notes to screen job applications changes its intended purpose and may become a provider under Article 25(1)(c), because analysing and filtering job applications is listed in Annex III point 4. With us, a new purpose is a new Vorgang anyway, not an extension of the existing one.
For Agentivo, operating means monitoring, correcting and taking responsibility
When we say we operate an AI employee, we describe our service after go-live: we monitor it, correct it and take responsibility for the result. In the AI Act, "operator" is an umbrella term for all the actors (Article 3(8), EUR-Lex), and the role with the day-to-day duties of use is the deployer.
Germany's Bundesnetzagentur explains (in German) that a legal person remains the deployer when it brings contractors into the operation of an AI system, as long as they act in its name and under its responsibility and control. Individual employees acting under the company's procedures are not deployers in their own right.
Who is the deployer and who is the provider for your Vorgang therefore follows from Articles 3 and 25 and from the contract, not from our choice of words. Settle it in writing before the agent starts.
Article 4 has required AI literacy measures since February 2025
AI literacy concerns every provider and deployer, whatever the risk of the system. Since the AI Omnibus, Article 4, EUR-Lex asks for measures that support the AI literacy of staff and other people who operate and use AI systems on the company's behalf, without requiring a specific level for any individual. The Bundesnetzagentur recommends documenting the measures and notes (in German) that the Act requires no certification of training.
For an operated agent, three groups matter most: the people who handle escalated cases, those who read the decision log, which is the agent's AI audit trail, and those allowed to trigger the intervention path. The article on EU AI Act Article 4 describes what they should know.
Article 50 has applied since August 2026 to systems that face people
According to the European Commission, the transparency rules have applied since August 2026. Providers ensure that people are informed when they interact with an AI system intended for direct interaction (paragraph 1) and mark synthetic content in a machine-readable way (paragraph 2). Deployers disclose, among other things, AI-generated text published to inform the public on matters of public interest, unless a person has reviewed it editorially and holds responsibility for it (paragraph 4).
Whether Article 50, EUR-Lex concerns a classification agent depends on whether it is intended to interact directly with people or generates content that is published. That is checked for each Vorgang, especially where the agent sends something to people, such as a receipt confirmation.
If the agent is high-risk, Article 26 adds deployer obligations from December 2027
Whether an agent is high-risk follows from Article 6 and Annex III and is assessed for each Vorgang; the criteria are explained in our article on EU AI Act compliance for a classification agent. If it is, the deployer obligations in Article 26, EUR-Lex apply to Annex III systems from 2 December 2027. This is how they divide in daily work:
| Obligation under Article 26 | What your company does | What operation by Agentivo contributes |
|---|---|---|
| Use in line with the instructions for use (26(1)) | uses the agent only for the agreed Vorgang | the agreed boundary and the handover document |
| Human oversight (26(2)) | names people with the necessary competence, training and authority | the confidence threshold below which the agent hands cases to them |
| Input data (26(4)) | ensures relevant and representative input data, where it controls them | classes and examples from the week-1 process conversation |
| Monitoring (26(5)) | informs provider and authority of a risk, suspends use and reports serious incidents | we monitor and correct; the intervention path stops the agent |
| Logs (26(6)) | keeps them for at least six months, where they are under its control | the decision log, with timestamp and reasoning |
| Informing workers (26(7)) | informs workers' representatives and affected workers before use | stays with you |
| Informing affected persons (26(11)) | informs them where the system makes or assists with decisions about them | stays with you |
The obligation stays with the deployer. The right-hand column shows what it can build on in operation, not that the obligation is met. Bodies governed by public law and private entities providing public services also assess the impact on fundamental rights before deployment (Article 27).
Articles 4 and 50 apply now, the high-risk obligations from December 2027
Regulation (EU) 2026/1744, EUR-Lex, the AI Omnibus, moved the dates for high-risk systems; according to the European Commission, it entered into force on 27 July 2026. For deployers, Article 113 in the consolidated text, EUR-Lex gives these dates:
| Since or from | What applies to deployers | Basis |
|---|---|---|
| 2 February 2025 | AI literacy | Article 4; Article 113(a) |
| 2 August 2026 | Transparency obligations for certain systems | Article 50; Article 113, second paragraph |
| 2 December 2027 | Deployer obligations for Annex III high-risk systems | Article 26; Article 113(c)(i) |
| 2 August 2028 | The same obligations for Annex I high-risk systems | Article 113(c)(ii) |
The original 2024 text had the Annex III obligations applying from 2 August 2026, so check older material against the consolidated text.
When someone asks, what counts is what is written down
Whether an auditor, an authority under Article 26(12) or your own management asks, a company that operates an AI agent should be able to show these records under the EU AI Act:
- the role allocation for the Vorgang, with a reference to the contract;
- the classification of the Vorgang under Article 6; for an Annex III system that is not considered high-risk, the provider documents it under Article 6(4);
- the AI literacy measures under Article 4;
- the decision log, with every assignment, timestamp and reasoning;
- the documented human intervention path;
- the data processing agreement under the GDPR; see GDPR compliant AI.
Every operation by Agentivo includes:
- Operation entirely within the EU, responsibility from Düsseldorf.
- GDPR-compliant processing with an AVV (data processing agreement) included.
- A complete decision log available as evidence for EU AI Act audits.
- A defined human intervention path, documented and triggerable at any time.
Agentivo is not a law firm. This article explains the obligations in the Act; whether they apply to your Vorgang and how you meet them is for you to assess with your legal advisers. Who is answerable for a running agent day to day is covered under AI agent governance.
Frequently asked questions
Does the EU AI Act apply to small companies?
Yes. Article 2 covers deployers of AI systems established or located in the Union, without a size threshold; only purely personal, non-professional use is excluded. Article 4 asks every deployer for AI literacy measures, and under Article 4(2) the Commission and Member States support SMEs in particular. Which further obligations apply depends on the company's role and the risk of the system.
Is our company the deployer if Agentivo operates the agent?
That follows from Article 3 and the contract, not from the word. The deployer is whoever uses the system under its own authority. According to the Bundesnetzagentur, a legal person remains the deployer when contractors work in its name, under its responsibility and control. For us, operating means monitoring, correcting and taking responsibility. Settle the roles for your Vorgang before the start, ideally in writing.
What applies if the agent is not high-risk?
Then the deployer obligations in Article 26 do not apply. Article 4 on AI literacy still applies, and Article 50 applies if the agent is one of the systems it covers. If a system is listed in Annex III but is not high-risk under Article 6(3), the provider documents that assessment before placing it on the market and registers the system under Article 49(2).
What applies already today?
Since 2 February 2025, Article 4: measures that support the AI literacy of the people working with the system. Since 2 August 2026, the transparency obligations in Article 50 for the systems it covers. The deployer obligations for Annex III high-risk systems follow from 2 December 2027 (consolidated text). Role allocation and classification can be prepared now, because both need facts from operation.
Does a Pilotbetrieb replace the legal review?
No. Agentivo is not a law firm. The Pilotbetrieb, our entry engagement, provides the facts a review rests on: the agreed Vorgang and its boundary, the decision log with timestamp and reasoning for every assignment, the documented intervention path and a written evaluation at the end. Whether obligations apply and are met is for you to assess with your legal advisers.
The next step
If your company assigns incoming documents by hand today and you want evidence for every assignment, that is the Vorgang for a Pilotbetrieb: one intake, one task, one destination. Fixed price, agreed before we start. Seven weeks, four of them productive operation on your real documents.