Introduction
The AI Act, the European regulation on artificial intelligence, already partly applies. For small and medium-sized enterprises (SMEs), the question is no longer whether to prepare, but what to check first. This article offers a simple compliance checklist; for the full timeline and the details of the obligations, read our guide to the AI Act for businesses.
At the end of 2026, three blocks of obligations are in place: prohibited practices (since 2 February 2025), the AI literacy measures required by Article 4, and the transparency rules (since 2 August 2026). The obligations on high-risk systems have been postponed by the Digital Omnibus to 2 December 2027. Training your teams is one of the first measures to take: our AI training for businesses is designed for exactly that.
Understanding the AI Act
The AI Act is a European regulation that entered into force on 1 August 2024 and classifies AI systems according to their level of risk. Companies must assess their systems to determine their classification and the resulting obligations. For a detailed overview of the risks, you can read our article on AI risk classification under the AI Act.
Step one: initial assessment
Identifying AI systems
Start by identifying all the AI systems used in your company. This includes internal applications, products offered to customers, third-party solutions and the tools that employees use on their own initiative.
Classifying systems
Classify each system according to the four levels of the AI Act: unacceptable risk (prohibited), high risk, limited risk (transparency obligations) and minimal risk. This classification determines the level of compliance required. Also note your role for each tool: most SMEs are deployers (they use a tool), not providers (they do not place it on the market).
Checking for prohibited practices
Make sure that no tool falls under the practices prohibited since February 2025, such as analysing employees' emotions at work or social scoring of individuals.
Step two: achieving compliance
Transparency
Since 2 August 2026, a person interacting with a chatbot must know that they are talking to an AI, unless it is obvious. Published deepfakes must be labelled as such.
Documentation
Make sure that each AI system comes with up-to-date documentation: what it is used for, what data it uses, who is responsible for it. For purchased tools, ask the provider for this documentation.
Impact assessment
For systems that will be high-risk from 2 December 2027 (recruitment, HR management, credit scoring, etc.), start preparing now an analysis of their effects on individuals, along with measures to limit the risks. If the tool is used to make decisions concerning your employees, plan to inform their representatives before it goes live.
Training and awareness
Train your employees in good practices for using AI. Since 27 July 2026, Article 4 requires companies to take measures to support the AI literacy of their teams: a usage policy and training tailored to each job are the simplest proof of this.
Step three: ongoing monitoring
Audit and review
Set up regular audit processes to assess the ongoing compliance of your AI systems. An internal audit can be a good starting point; for more details, read our article on how to assess your company's AI maturity.
Reports and documentation
Keep records of all the assessments and audits carried out. These documents may be requested by the authorities in the event of an inspection.
Additional resources
For SMEs that want to get the most out of their AI investment while remaining compliant, it can be useful to know how to calculate the return on investment of these technologies. See our guide to AI ROI. If you process personal data, also read GDPR and AI: how to stay compliant.
Frequently asked questions
What is the AI Act?
The AI Act is European Regulation (EU) 2024/1689, which governs the use of AI systems to ensure that they are safe and respect fundamental rights. It has been applying in stages since 2025.
How do I classify my AI systems?
Classification is based on the level of risk a system presents: unacceptable, high, limited or minimal. Recruitment and credit scoring, for example, are high-risk, a chatbot falls under transparency, and a spell checker is minimal risk.
What are the risks of non-compliance?
Fines can reach €35 million or 7% of worldwide annual turnover for a prohibited practice. For an SME, the lower of the two amounts is the cap. Reputational risk comes on top of that.
💡 Did you know? diagonall is a French AI agency that supports companies in their AI transformation with training, strategic consulting and custom solutions.
Ready to take action?



