Knowledge base
AI Act checklist for SMEs: what to sort out in 7 steps
By Bob Biesemans · published on
A practical checklist for SMEs that use AI tools: prohibited AI, high risk, data, AI literacy, labelling and your register. Status November 2026.
Short answer
An SME that uses AI tools takes seven steps: map the AI in use, check that none of it is prohibited, see whether anything is high-risk, agree rules on data, take measures for AI literacy, label where Article 50 requires it, and record everything in a register that you keep up to date.
Who this checklist is for
This checklist is for an SME that uses AI but doesn't build it: staff work with ChatGPT, Microsoft 365 Copilot, Gemini or with AI built into accounting, CRM or HR software. Under the AI Act you are then a deployer. Most obligations for AI providers don't apply to you.
1. Map the AI you use
List the AI tools in your company, per department and per task. Don't forget the AI features inside existing software, or the private accounts staff use on their own. Without that list you can't take the next steps.
2. Check that you use no prohibited AI
Since 2 February 2025 some practices have been prohibited (Article 5), such as harmful manipulation, social scoring and inferring employees' emotions in the workplace, except for medical or safety reasons. For most SMEs this is quickly ticked off, but look closely at tools that analyse staff voices, faces or behaviour.
3. Check whether anything is high-risk
Annex III of the AI Act lists high-risk uses, such as AI that filters CVs or ranks candidates, monitors employee performance or assesses people's creditworthiness. Under the Digital Omnibus, the obligations for deployers of such systems apply from 2 December 2027: use according to the instructions, human oversight and informing staff beforehand. Note now which tools fall under this.
4. Agree rules on data
The GDPR fully applies to AI. Agree which data may go into which tool: public information in the approved tools, internal information only in a business environment with agreements, and no personal data, contracts or passwords unless the company has explicitly provided a safe solution for it.
5. Take measures for AI literacy
Article 4 asks for measures for the AI literacy of those who use AI on the company's behalf, tailored to their knowledge and their work. There is no mandatory training or certificate. Measure per department how staff use AI in practice, explain where needed and measure again.
6. Label where required
Since 2 August 2026 the transparency rules of Article 50 apply. Realistic AI-generated images, video or audio of real people or events (deepfakes) need a disclosure. So do texts published to inform the public on matters of public interest, unless a person reviewed them and holds editorial responsibility. A quote or customer email you checked yourself usually isn't covered.
7. Record everything and repeat
Keep a simple register: which tools, which rules, which measures for whom and when, where the evidence is and the next step. Repeat the exercise when new tools arrive. In Belgium the federal coalition agreement names the BIPT as the AI Act supervisor; for personal data the Data Protection Authority is competent. General information, not legal advice.
Source: AI Act, (EU) 2024/1689 (EUR-Lex)
Frequently asked questions
Which AI Act rules already apply to an SME?
The prohibited practices and Article 4 on AI literacy have applied since 2 February 2025, the transparency rules of Article 50 since 2 August 2026. Under the Digital Omnibus, the obligations for deployers of Annex III high-risk AI follow on 2 December 2027.
Does my SME use high-risk AI?
Usually not. Everyday writing and search tasks with ChatGPT or Copilot are not covered. AI that filters CVs or ranks candidates, monitors employee performance or assesses people's creditworthiness is.
Who supervises in Belgium?
The federal coalition agreement names the BIPT as the AI Act supervisor. Where AI processes personal data, the Data Protection Authority is competent.