Guide
The EU AI Act for small businesses: what changes if you use an agent
By Ruslan Rio · Updated · 4 min read
For a small business that uses an AI agent in its processes, the EU Artificial Intelligence Act mainly asks for three things: that anyone talking to the agent knows it is an AI, that the team using it is trained, and that it is not used to decide about people without the safeguards for high-risk systems. Most business agents — invoices, email, quotes, reports — are not high-risk. This guide sums up what applies now and what comes later.
An explainer, not legal advice. Dates checked on 26 October 2026.
Provider or deployer
The Act assigns obligations by role. The provider is whoever develops an AI system, or has one developed, and places it on the market or puts it into service under its own name. The deployer is whoever uses it in their professional activity.
A company that uses an agent in its own process is a deployer. If it also commissions a custom agent and puts it into service under its own name, the Act may treat it as a provider too. It is worth setting out in the contract who takes on each role.
What already applies
- Prohibited practices, since 2 February 2025: among others, harmful manipulation, social scoring and emotion recognition in the workplace. An invoice or email agent comes nowhere near any of them.
- AI literacy (Article 4): whoever uses AI systems must take steps so their staff have the knowledge needed to use them. The Digital Omnibus reworded its scope, but it remains. In practice it is met by training the team that works with the agent, which is part of any serious implementation.
- Transparency (Article 50), since 2 August 2026: if a system converses with people, it must be clear that it is an AI unless that is obvious. This applies squarely to phone receptionists and customer-service assistants. Content generated or manipulated with AI, such as deepfakes, must also be labelled.
High-risk, postponed
High-risk systems are those the Act lists for their impact on people — recruitment and staff management, assessing the creditworthiness of individuals, access to essential services or education, among others (Annex III) — and those that are part of regulated products (Annex I). They carry demanding obligations: risk management, documentation, record-keeping and human oversight.
The Digital Omnibus (Regulation (EU) 2026/1744, in force since 27 July 2026) postponed those obligations: Annex III applies from 2 December 2027 and Annex I from 2 August 2028.
Most small-business agents are not high-risk. But an agent that screens candidates or decides on an individual’s credit would be, and it is worth designing it that way from the start even though the obligation arrives later.
Timeline
What to do if you use, or plan to use, an agent
- Take stock: which AI systems you use and what for.
- Check whether any of them decides about people — employment, credit, access to services. If so, treat it as high-risk.
- If the agent talks to customers, in writing or by phone, have it say it is an AI.
- Train whoever works with it: what it does, what it does not do and when it needs reviewing.
- Write down what the agent decides and what a person decides.
- Remember that the GDPR still applies to any personal data the agent handles.
In Spain the authority supervising the Act is the Spanish Agency for the Supervision of Artificial Intelligence (AESIA). In our diagnosis, the process map states which category of the Act each case falls under, before anything is built.
Shall we look at your process?
The diagnosis is free: we measure where your team’s time goes and tell you what is worth automating.
Start the conversation