# EU AI Act Deployer Obligations for SMEs: Duties and Dates

> What the EU AI Act asks of a company that uses AI: Articles 4, 26, 27, 50 and 86, who each one binds and from when. Checked on EUR-Lex, 2026-10-08.

- Published: 2026-11-03
- Author: Jacobo González Jaspe
- Language: en
- Tags: EU AI Act, Deployers, Article 26, SMEs, Compliance
- Canonical URL: https://vorluxai.com/en/blog/eu-ai-act-deployer-obligations-smes-en/
- Source: VORLUX AI (J4SGON S.L.), https://vorluxai.com/

---

By the end of this guide you will know which EU AI Act duties your company carries as a user of AI, which ones apply today and which only start in December 2027. You need a spreadsheet and an honest hour, nothing else.

> **TL;DR**
> * Almost every SME that uses ChatGPT-style tools, a chatbot or an image generator is a **deployer**. Two duty sets apply now: AI literacy (since 2 February 2025) and the transparency notices of Article 50(3)-(4) (since 2 August 2026).
> * The heavy list, the deployer obligations for high-risk systems, only applies if you use AI for things like screening CVs or scoring credit. The Digital Omnibus moved it to **2 December 2027**.
> * Verdict: **Do it now** for the inventory and the training; **plan, do not panic** for the high-risk list.

## Step 0: confirm that you are a deployer

The AI Act ([Regulation (EU) 2024/1689](https://eur-lex.europa.eu/eli/reg/2024/1689/oj)) splits companies into roles. Article 3(4) defines a deployer as anyone who uses an AI system under its authority, except for personal, non-professional activity. If your staff use an AI tool for work, your company is its deployer.

One trap is worth knowing. Article 25(1) turns a deployer into a provider of a high-risk system when it puts its own name on that system, makes a substantial modification to it, or changes the intended purpose of a general tool so that it becomes high-risk. Wiring a general chatbot into your hiring process is the classic way to do that by accident.

## The duties and their dates

We read every row below in the consolidated text on EUR-Lex and in the [Digital Omnibus on AI, Regulation (EU) 2026/1744](https://eur-lex.europa.eu/eli/reg/2026/1744/oj), which entered into force on 27 July 2026. Checked on 2026-10-08.

| Provision | What it asks | Who it binds | Applies from |
|---|---|---|---|
| Art. 4 | Take measures to support the AI literacy of staff who use AI | Every provider and deployer | 2 February 2025 (wording amended by 2026/1744) |
| Art. 50(3) | Tell people when emotion recognition or biometric categorisation runs on them | Deployers of those systems | 2 August 2026 |
| Art. 50(4) | Disclose deep fakes, and AI text published to inform the public on matters of public interest | Deployers | 2 August 2026 |
| Art. 26 | Use per instructions, human oversight, logs, monitoring, informing workers and people | Deployers of high-risk systems | 2 December 2027 (Annex III); 2 August 2028 (Annex I) |
| Art. 27 | Fundamental rights impact assessment before first use | Public bodies, private entities providing public services, credit and life/health insurance scoring | Same as Art. 26 |
| Art. 86 | Explain the role of AI in a decision that significantly affects a person | Deployers of Annex III high-risk systems | General date, 2 August 2026 |

The dates for the high-risk rows come from Article 113, third paragraph, point (c), as replaced by the Omnibus. The [Commission's AI Act page](https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai) gives the same two dates.

## Step 1: write the inventory

Open a spreadsheet with one row per AI use, not per tool. A writing assistant used for both marketing and HR replies is two rows. Columns that make the next steps easy:

1. Tool and vendor
2. Task it does, in one sentence
3. Who uses it and who sees the output (staff, customers, the public)
4. Personal data it touches
5. Does it help decide something about a person? (yes/no)

Your AI assistant can sort the rows for you if you give it the right instruction:

```text
Here is my company's list of AI uses (one per line). For each one, say whether it
could fall under Annex III of Regulation (EU) 2024/1689 (for example recruitment,
worker evaluation, creditworthiness, life or health insurance pricing) and why.
Answer "unclear" when the description is not enough. Do not invent obligations.
```

Decide the final answer yourself. A model that is wrong here costs you either wasted work or a missed duty.

## Step 2: look for high-risk uses

For a typical small company the Annex III list comes down to a few questions:

* **Hiring.** Do you use AI to place targeted job ads, filter applications or evaluate candidates (Annex III, point 4(a))?
* **Managing staff.** Do you use AI to decide on promotions, terminations or task allocation, or to monitor performance (point 4(b))?
* **Credit and insurance.** Do you score the creditworthiness of people, or price life and health insurance (points 5(b) and 5(c))?

If every answer is no, the high-risk list does not apply to you. Write that down with the date. It is the most useful line in the whole file.

## Step 3: if you found one, prepare the deployer obligations

The deployer obligations in Article 26 are concrete. Here they are in plain words, with the paragraph so you can check the text:

| Paragraph | What you do |
|---|---|
| 26(1) | Use the system according to the provider's instructions for use |
| 26(2) | Name the people who oversee it; they need competence, training and authority |
| 26(4) | Where you control the input data, keep it relevant to the purpose |
| 26(5) | Monitor it; if it presents a risk, inform the provider and the authority and suspend use |
| 26(6) | Keep the logs under your control for at least six months, unless other law says otherwise |
| 26(7) | As an employer, inform workers' representatives and affected workers before using it at work |
| 26(9) | Use the provider's information for your GDPR data protection impact assessment |
| 26(11) | Tell people that a high-risk system helps make decisions about them |

Paragraph 26(8) adds registration duties for public authorities. Article 27(1) adds the fundamental rights impact assessment for public bodies, private entities providing public services, and deployers who score credit or price life and health insurance.

Since the Omnibus, Article 27(4) lets that assessment point to the relevant sections of a data protection impact assessment you already have, so the work is not done twice.

Then there is Article 86(1): a person affected by a decision based on an Annex III system can ask you for a clear explanation of the role the AI played. Prepare a short template answer per system now, while you still remember how it works.

## Step 4: the duties that apply today

These are the ones to finish this month.

* **AI literacy.** Article 4 on AI literacy, as reworded by the Omnibus, asks you to take measures to support the literacy of the people who use AI on your behalf. It does not require you to guarantee a specific level for each person. A one-hour session per team, a usage policy and a record of who attended is a proportionate start. Our [AI literacy training plan](https://vorluxai.com/en/blog/eu-ai-act-article-4-ai-literacy-training-plan-en/) has a template.
* **Transparency notices.** The Article 50 transparency rules split between provider and deployer. Our [Article 50 guide](https://vorluxai.com/en/blog/eu-ai-act-article-50-label-your-chatbot-int-50-en/) lists the labels to copy for chatbots, images, voice and text.

## When this is not your problem

* **Personal use.** A freelancer using an AI tool for private matters is outside the deployer definition.
* **No Annex III use.** Then Article 26 and Article 27 never reach you, whatever the date. The inventory is still worth keeping.
* **The legal question itself.** This guide is a reading aid for the regulation, not legal advice. Spanish law may add rules, and our [AESIA guide](https://vorluxai.com/en/blog/aesia-what-spanish-businesses-must-know-2026-en/) explains the role of Spain's AI supervision agency.

## The one-page checklist

- [ ] Inventory of AI uses, one row per task, dated
- [ ] Role per row: deployer, or provider because of Article 25(1)
- [ ] Annex III check with a written "no" or a list of systems
- [ ] Literacy measures recorded: session, policy, attendance
- [ ] Article 50 notices in place where people meet AI output
- [ ] For each high-risk system: overseer named, logs kept six months, workers informed
- [ ] Explanation template per high-risk system, ready for Article 86 requests
- [ ] Next review date in the calendar

## Next steps

* Run the [EU AI Act self-assessment scorecard](https://vorluxai.com/en/blog/eu-ai-act-self-assessment-scorecard-en/) to see where you stand.
* Use the [EU AI Act page](https://vorluxai.com/en/eu-ai-act/) checklist; it takes about three minutes.
* For Article 26(9), see [when the GDPR requires a DPIA and how to run one for local AI](https://vorluxai.com/en/blog/gdpr-article-25-local-ai-privacy-by-design-2026-en/).

## Work with us

If you would rather do the inventory and the training with someone who has read the text, our entry offer "EU AI Act: deployer-ready in two weeks" covers exactly that. It is described on the [EU AI Act page](https://vorluxai.com/en/eu-ai-act/#oferta). [Book a free 15-minute call](https://vorluxai.com/en/contact/) or see [how we work in consulting](https://vorluxai.com/en/consulting/).
