# The EU AI Act for Companies That Use AI: What Applies in 2026
- **url:** https://easify-ai.com/eu-ai-act-for-companies-that-use-ai-2026/
- **published:** 2026-11-16
- **topics:** ai-automation-in-business
- **publisher:** Easify AI, part of Easify Technologies
Most companies are not building AI. They use Microsoft 365 Copilot, ChatGPT, a support chatbot or an automation that drafts replies. The EU AI Act still applies to them, but in a much narrower way than the headlines suggest.
This guide is for owners and operations leads who want to know what applies now, what is coming, and what changed when the EU delayed parts of the Act in 2026. It reflects the position as of November 2026. It is not legal advice, and some guidance from the Commission is still being written.
## Are you a provider or a deployer?
The AI Act gives different duties to different roles. The two that matter for most businesses:
- **Provider:** develops an AI system or model, or has one developed, and puts it on the market under its own name. OpenAI, Anthropic, Microsoft and the vendor of your HR screening tool are providers.
- **Deployer:** uses an AI system under its own authority in a professional context. A company that rolls out Copilot, or uses an AI tool to sort support tickets, is a deployer.
Most of the Act's obligations sit with providers. Deployers have a shorter list, which grows only when the use is high-risk.
> **Watch out:** you can become a provider without planning to. Putting your brand on an AI system, substantially modifying one, or using a general-purpose tool for a high-risk purpose can shift provider duties onto you. A chatbot on your website built on a model API is the case to check first.
## The timeline, after the AI Omnibus
The AI Act entered into force on 1 August 2024 and applies in stages. In November 2025 the Commission proposed a "Digital Omnibus on AI" to delay the high-risk rules while standards and guidance catch up. Parliament and Council adopted it in June 2026, and it was published as Regulation (EU) 2026/1744, entering into force on 27 July 2026.
| Date | What applies | Status |
|---|---|---|
| 2 February 2025 | Prohibited practices; AI literacy (Art. 4) | In application |
| 2 August 2025 | Rules for general-purpose AI models; governance; penalties | In application |
| 2 August 2026 | General application, including most transparency duties (Art. 50) | In application |
| 2 December 2026 | New ban on AI generating child sexual abuse material or non-consensual intimate images; end of grace period for machine-readable marking by providers of systems already on the market | Coming |
| 2 August 2027 | General-purpose models placed on the market before August 2025 must comply | Coming |
| 2 December 2027 | High-risk obligations for Annex III uses (e.g. employment, education, credit, biometrics), including deployer duties | Delayed from August 2026 |
| 2 August 2028 | High-risk AI embedded in regulated products (Annex I, e.g. machinery, medical devices) | Delayed from August 2027 |
> **Note:** the delay is a postponement, not a cancellation. The high-risk duties are unchanged in substance, and the Commission's guidelines on classifying high-risk systems are still due. Plan for December 2027, and check the AI Act Service Desk for updates.
## What applies to you now
### Prohibited practices (since February 2025)
Article 5 bans a short list of uses outright. For ordinary businesses the relevant ones are:
- **Emotion recognition in the workplace** or in education, except for medical or safety reasons.
- **Manipulative or deceptive techniques** that distort behaviour and cause significant harm.
- **Exploiting vulnerabilities** linked to age, disability or social and economic situation.
- **Social scoring** that leads to unjustified detrimental treatment.
- **Biometric categorisation** to infer sensitive traits such as political views or sexual orientation, and untargeted scraping of facial images.
Few companies do any of these on purpose. The risk is in bought tools: a call-centre product that scores agents' emotions, for example. Fines for prohibited practices go up to €35 million or 7% of worldwide annual turnover.
### AI literacy (Article 4)
Since February 2025, providers and deployers have had to address the AI literacy of staff who operate or use AI systems on their behalf. The Omnibus softened the wording: the duty is now to take measures that support AI literacy, taking into account people's technical knowledge and the context of use, without guaranteeing any specific level for each person.
In practice the advice has not changed. Train people on the tools they use, the risks of those tools (wrong answers, confidential data, bias), and your internal rules. Keep a record of who was trained on what.
### Transparency (Article 50, since August 2026)
Most transparency duties applied from 2 August 2026. Some sit with providers, some with deployers:
| Duty | Who | What it means in practice |
|---|---|---|
| Tell people they are talking to AI | Provider | Chatbots must say they are AI, unless it is obvious. If you built and branded the bot, that may be you. |
| Mark synthetic content as AI-generated | Provider | Machine-readable marking; providers of systems already on the market have until 2 December 2026. |
| Inform people exposed to emotion recognition or biometric categorisation | Deployer | Where these uses are allowed at all, people must be told. |
| Disclose deepfakes | Deployer | Images, audio or video that look real and were generated or altered by AI must be labelled, with lighter rules for clearly artistic or satirical work. |
| Disclose AI-generated text published to inform the public | Deployer | Unless a person has reviewed it and someone holds editorial responsibility. |
> **Tip:** the editorial exemption is worth noticing. A blog post or press release drafted with AI, then reviewed and approved by a named person, does not need an AI label under Article 50. Keep a simple record of who approved it.
## High-risk uses: what deployers will have to do
High-risk uses are listed in Annex III. The ones businesses meet most often are **recruitment and HR decisions** (screening CVs, ranking candidates, deciding promotions or terminations), **creditworthiness checks**, and **life and health insurance pricing**. Using Copilot to draft emails is not high-risk. Using an AI tool to shortlist job applicants is.
From 2 December 2027, deployers of high-risk systems must, under Article 26:
1. **Use the system according to the provider's instructions**, with appropriate technical and organisational measures.
2. **Assign human oversight** to people who have the competence, training and authority to do it.
3. **Make sure input data you control is relevant** and sufficiently representative for the purpose.
4. **Monitor the system** and inform the provider and authorities of serious incidents or risks.
5. **Keep the logs** the system generates, for at least six months unless other law says otherwise.
6. **Inform workers and their representatives** before using a high-risk system in the workplace.
7. **Inform people** when a high-risk system is used to make or help make decisions about them.
Public bodies, private companies providing public services, and firms using AI for credit scoring or life and health insurance pricing must also complete a fundamental rights impact assessment (Article 27).
## General-purpose AI: mostly the provider's problem
The rules for general-purpose AI models, in application since August 2025, require providers such as OpenAI, Anthropic, Google and Mistral to publish technical documentation, a copyright policy and a summary of training content. Models with systemic risk carry extra duties.
As a business user you have nothing to file. What you should do is keep the provider documentation and terms on record, because they will feed your own risk assessment if you ever build something higher-risk on top of a model.
## A practical plan for the next 12 months
```flow
trigger | AI inventory | Every tool, who uses it, for what | Excel
ai | Classify each use | Prohibited, high-risk, transparency or minimal
human | Owner signs off | Decisions on risky uses and training plan
tool | Train by role | Tools, risks and your rules | Microsoft Teams
output | Register kept current | Reviewed every quarter | SharePoint
```
1. **List every AI tool in use**, including features switched on inside software you already pay for.
2. **Classify each use**: prohibited, high-risk, transparency duty, or minimal risk. Most will be minimal.
3. **Write a one-page AI use policy**: approved tools, what data may go in, who reviews outputs.
4. **Train staff by role**, and record it.
5. **Label where required**: chatbots, synthetic images and published AI text without editorial review.
6. **Flag any high-risk use** (HR screening is the usual one) and ask the vendor how they will meet the December 2027 deadline.
Training is where most of the effort sits. Our [Microsoft 365 Copilot rollout](/use-cases/microsoft-copilot-rollout/) includes role-based training and usage rules, which covers much of the literacy duty for teams starting with Copilot.
## Summary
- **You are probably a deployer**, with lighter duties than a provider, unless you brand or substantially change an AI system.
- **Already in force:** prohibited practices and AI literacy (February 2025), and most transparency duties (August 2026).
- **Delayed:** high-risk obligations now start on 2 December 2027, or 2 August 2028 for AI in regulated products.
- **Watch HR tools:** recruitment and employee decisions are the most common high-risk use in ordinary businesses.
- **Act now on three things:** an inventory of AI use, role-based training, and labelling where AI talks to or about people.
- **Still moving:** Commission guidelines are pending, so check official sources before relying on any date.
## FAQ
### Does the EU AI Act apply to a company that only uses ChatGPT or Copilot?
Yes, but lightly. You are a deployer. For everyday uses such as drafting, summarising and searching, the main duties are AI literacy for staff and avoiding prohibited practices. Heavier duties only apply if you use AI in a high-risk area such as hiring or credit decisions.
### Were the high-risk rules delayed?
Yes. The AI Omnibus, published in the Official Journal in July 2026, moved obligations for high-risk systems listed in Annex III to 2 December 2027, and for high-risk AI built into regulated products to 2 August 2028.
### Is AI literacy training still mandatory?
The duty remains, but the Omnibus softened its wording. Providers and deployers must take measures to support the AI literacy of staff who use AI, rather than guarantee a particular level. A documented, role-based training programme is still the sensible way to show you have done this.
### Can we become a provider without meaning to?
Yes. If you put your own name on an AI system, substantially modify one, or change its purpose so that it becomes high-risk, you can take on provider obligations. Building a customer-facing chatbot under your brand is the most common case to check.