Regulation
EU AI Act compliance: what your business needs to do
EU AI Act compliance also concerns companies that simply use off-the-shelf AI tools, and the obligations arrive in phases. This guide explains what changes for a small business, how to classify each use of AI with three questions and which steps to take now.
What the EU AI Act is and when it applies
The managing director of an estate agency in Dublin gets an email from her CRM provider about EU AI Act compliance. Her negotiators use ChatGPT to write listings, the website has an AI chat and HR recently trialled a tool that ranks CVs. She wants to know what is actually expected of her.
The EU AI Act is Regulation (EU) 2024/1689 on artificial intelligence. It is the European Union's first broad law on AI and, like the GDPR, it applies directly in every member state. It can also reach businesses outside the EU, such as UK firms, when they offer AI systems in the EU or the output is used there.
The Act entered into force in August 2024 and its obligations arrive in phases. Bans on certain practices and the AI literacy duty in Article 4 have applied since February 2025. AI literacy means each person knows enough to use AI sensibly in their role. The adopted timeline sets August 2026 as the general date for most other obligations, including many transparency and high-risk rules.
The EU has since discussed postponing some of those obligations, mainly the high-risk ones. Before you set internal deadlines, check the current official timeline on the European Commission's website or with your adviser.
Most of the Act targets those who build AI systems, which it calls providers. It also sets duties for companies that use AI systems in their business, which it calls deployers. The Dublin agency is in this second group.
EU AI Act requirements by risk level
The Act sorts uses of AI by the risk they pose to people. The higher the risk, the stricter the rules.
Prohibited practices
Some uses are banned outright. They include social scoring, techniques that manipulate people or exploit their vulnerabilities, emotion recognition in the workplace or in education except for medical or safety reasons, and untargeted scraping of facial images to build recognition databases.
High-risk systems
These are uses with a serious impact on people's lives, such as recruitment, employee evaluation, access to credit or education. Deployers must follow the provider's instructions for use, keep human oversight in place and, at work, inform the employees affected.
Transparency obligations
Someone chatting with a chatbot should know they are not talking to a person, unless that is obvious. Generated or manipulated images, audio or video that look real must be labelled as such.
Minimal risk
Most everyday uses in a small business, such as drafting emails, summarising documents or preparing slides, carry no specific obligations under the Act beyond AI literacy. The GDPR, confidentiality duties and other laws still apply. The rules for the large models behind ChatGPT or Gemini fall on the companies that build them.
A three-question triage: which category does each AI use fall into?
Category lists make sense in the abstract and are harder to apply. Ask these three questions, in this order, about every tool your team uses. Each tool takes a few minutes.
- 01
Does it decide on, filter or score people?
If the tool ranks candidates, rates staff performance or affects someone's access to credit or education, mark it red. It may be high-risk and is worth reviewing with an adviser before you keep using it.
- 02
Does a customer interact with it or see its output as real?
A website chat, a WhatsApp assistant or an AI-edited photo that looks genuine is amber. Transparency duties apply, such as saying that AI is involved or labelling generated content.
- 03
Does it only help an employee do their work?
Drafting, summarising, translating or preparing slides that a person checks is green. Article 4 training is enough, plus the GDPR if personal data is involved.
- 04
Check whether you act as a provider
If your company builds its own system or markets one under its name, it may take on provider obligations, which are stricter. That is rare for a small business using off-the-shelf tools, but worth a conversation with an adviser if it applies.
EU AI Act compliance for a small business in 6 steps
If your company uses off-the-shelf AI tools and does not build high-risk systems, these steps cover the essentials. Most businesses can work through them in a few weeks.
- 01
List every use of AI
Write down which tools are used, by whom, for what and with which data. Include AI features built into software you already have, such as your CRM, email or accounting package. A spreadsheet with one row per use is enough.
- 02
Triage each row
Add a column with the colour that comes out of the three questions. Review red rows with an adviser first. Look closely at HR.
- 03
Be transparent where it is required
If your website or WhatsApp channel has an AI assistant, say so clearly when the conversation opens. If you publish generated images or videos that could pass for real, label them next to the content.
- 04
Train your team and keep a record
Training should match each role. A salesperson drafting proposals with AI needs to know how to check the output and which data must stay out. Record who received which training and when.
- 05
Write a usage policy
A short document sets out which tools are approved, which data stays out and who reviews the output. Keep it to two pages so people actually read it.
- 06
Review your vendors' terms
Read each tool's terms of use, data processing terms and anything the provider publishes about the AI Act. Check whether your inputs are used for training and where data is processed. Get it in writing for any tool your team uses with client data.
Example
A Dublin estate agency with twelve negotiators maps its AI use
Back to the Dublin agency, a typical case with twelve negotiators. The managing director spends a morning on the list with the office manager. Seven uses come up, including listing copy written with ChatGPT, photos of flats furnished virtually with AI, the website chat, automatic call summaries in the CRM and the CV tool HR trialled.
The triage sorts the work. The CV tool comes out red, so they pause it and ask their adviser before deciding whether and how to use it. The virtually furnished photos and the chat come out amber. From that week, every listing with virtual furniture says so under the photo, and the chat introduces itself as an AI assistant in its first message. Everything else is green.
For Article 4, they run practical training by role, one session for negotiators and one for the office team, and keep the attendance list with dates. The two-page usage policy bans entering clients' identity or bank details into unapproved tools.
To track progress, the director checks that the list is complete, with an owner on every row, that the whole team has been trained and that no red use is running without review. She repeats the list every six months, because software adds AI features without much notice.
The AI Act and the GDPR apply side by side
The AI Act does not replace the GDPR. If an AI tool processes personal data about clients or staff, your company has to comply with both. The AI Act looks at the system and its level of risk, while the GDPR looks at people's data and how it is handled.
In practice, the work overlaps. Your list of AI uses is a good starting point for updating your record of processing activities, and one training session can cover both topics.
National authorities supervise compliance, and the Act provides for fines that can be substantial, especially for prohibited practices. For a small business, the main risk usually lies in not knowing what is being used or with which data, in forgetting the AI features inside familiar software, or in assuming the law only concerns large companies.
A spreadsheet and three questions to get started
Open a spreadsheet and ask each team lead to list the AI tools their people use. Apply the three questions to every row and book time with your adviser only for the red ones. Add an AI notice to your chat if you have one, and set a date for training.
This page is general information and is not legal advice. Check your own situation with your adviser and confirm the current official timeline before you set deadlines.
Obinova is an AI agency that helps companies put AI to work for their people. Our training on ChatGPT, Claude, Gemini and Microsoft Copilot starts from concrete use cases in your team's work and is a practical way to build the AI literacy that Article 4 asks for. It is delivered on a dedicated video platform or in person at your company.
Frequently asked questions
Does the EU AI Act apply to a small business that only uses ChatGPT?
Yes, though in a limited way. If your company uses off-the-shelf AI tools to draft or summarise, your main duties are to train staff under Article 4, avoid prohibited practices and meet the transparency rules where they apply. The strictest requirements fall on those who build or use high-risk systems. Check your own case with an adviser.
When did AI training for employees become mandatory?
Article 4 of the AI Act, on AI literacy, has applied since February 2025. It asks companies that use AI systems to take measures so their staff have a sufficient level of AI knowledge, based on their role and how the systems are used. It does not set a number of hours or a specific format. Keeping a record of the training is sensible.
Do I have to say that the chatbot on my website is AI?
The Act requires that people know they are interacting with an AI system, unless it is obvious from the context. This duty is among those scheduled for August 2026 in the adopted timeline, so check the current official date. A clear message when the chat opens is usually enough, and it also builds trust with customers.
Is using AI for recruitment high-risk?
AI systems used to recruit or select people, for example to filter CVs or assess candidates, are listed among the high-risk areas in the Act. If your company uses such a tool, it will have its own duties as a deployer, such as keeping human oversight and informing staff. Review the case with an adviser before you introduce the tool or carry on using it.
What fines does the EU AI Act set for small businesses?
The Act sets fines according to the type of breach, with the highest amounts reserved for prohibited practices. When fines are set, the size of the company and the situation of small businesses are taken into account. National authorities handle supervision. For a small business using common tools, a list of AI uses and documented training greatly reduce exposure.
What does EU AI Act compliance cost a small business?
There is no fee to pay. The cost is mostly your team's time to list AI uses, train staff and write a usage policy, plus an adviser's review for any red use. It depends on how many tools you use and whether anything touches recruitment or credit. Check the current official timeline before setting deadlines, as the EU has discussed moving some dates.