The EU AI Act for SMEs: Compliance & AI Governance
The AI Act is not just for tech companies and multinationals. As an SME that uses AI tools, you are already subject to new obligations.
Updated: November 2026
Revised now that the Digital Omnibus is final. On 19 November 2025 the European Commission proposed a 'Digital Omnibus on AI'; the European Parliament voted in favour on 16 June 2026 and the Council followed on 29 June 2026. The amending regulation, Regulation (EU) 2026/1744, was published in the Official Journal on 24 July 2026 and has been in force since 27 July 2026.
What changes: the heaviest high-risk obligations move to December 2027 and August 2028, the AI-literacy obligation (Article 4) has been reworded into a duty to take measures, and a new prohibition applies from 2 December 2026. What does not change: for most businesses the AI Act is not delayed. Article 4 has applied since February 2025 and the transparency rules in Article 50 since 2 August 2026. The timeline below has been updated accordingly.
What Is the AI Act?
The AI Act (officially: EU Regulation 2024/1689) is the first comprehensive European law regulating the use of artificial intelligence. The regulation entered into force on 1 August 2024 and, like the GDPR, applies directly in all EU member states without requiring national transposition.
Its aim is twofold: to protect citizens from high-risk AI applications, while leaving room for innovation through a clear and predictable framework. To achieve this, the legislator chose a risk-based approach: the higher the potential risk, the heavier the obligations.
The AI Act takes effect in phases. Not all obligations apply immediately. Some have applied since 2025 and 2026, and the rest follow in 2027 and 2028. That makes it straightforward to look at what is already in play and what is coming, step by step.
Disclaimer: this article provides an informational overview and does not constitute legal advice. For guidance on how the AI Act applies to your specific situation, consult a lawyer or compliance specialist.
The Risk-Based Approach: Four Tiers
The AI Act classifies AI systems into four risk categories. Which category applies determines which obligations you face. Here is a concise overview.
Prohibited AI Practices
Certain AI applications are outright banned in the EU (Article 5): systems that pose an unacceptable threat to fundamental rights or human dignity. The main ones:
- Social scoring: rating people on their behaviour or personal characteristics in a way that leads to detrimental or disproportionate treatment. The ban covers public authorities and companies alike.
- Real-time remote biometric identification in publicly accessible spaces for law enforcement (with a few narrowly defined exceptions).
- AI that manipulates behaviour through techniques people are not aware of, or that exploits vulnerabilities such as age, disability or a precarious financial situation.
- Untargeted scraping of facial images from the internet or CCTV footage to build facial-recognition databases.
- Emotion recognition in the workplace and in education, except for medical or safety reasons.
- New under the Digital Omnibus, from 2 December 2026: AI systems that generate child sexual abuse material or non-consensual intimate imagery (such as so-called nudify apps).
High-Risk AI Systems
High-risk systems are permitted, but carry heavy requirements: conformity assessment, registration in an EU database, technical documentation, a risk-management system, and mandatory human oversight. This applies to AI used in sectors such as medical devices, biometric identification, critical infrastructure, education, and HR.
- Recruitment screening and CV evaluation via AI.
- AI systems used in credit scoring or insurance underwriting.
- AI in courts, border control, or policing.
- Medical diagnosis or treatment support systems.
Limited Risk: Transparency Obligations
AI systems in this tier face lighter obligations, primarily around transparency (Article 50). People must know they are dealing with AI. These rules have applied since 2 August 2026. Who has to do what is covered in labelling AI content under the AI Act.
- Chatbots and virtual assistants: make clear that the user is talking to AI, unless that is already obvious.
- Deepfakes (images, audio or video that look real): disclose that they were generated or manipulated by AI.
- AI-generated text published to inform the public on matters of public interest: disclose it, unless a person has reviewed the text and takes editorial responsibility for it.
- Emotion recognition or biometric categorisation: inform the people you use it on. Note that emotion recognition in the workplace and in education is prohibited.
- Providers of generative AI must also mark their output as AI-generated in a machine-readable format. For systems already on the market before 2 August 2026, that applies from 2 December 2026.
Minimal Risk
The vast majority of AI applications fall here: spam filters, AI-powered search, product recommendations, grammar tools. No specific AI Act obligations apply, though other laws (GDPR, sector regulation) continue to apply.
What Is Already in Effect and What Is Coming?
The AI Act has a phased implementation schedule. Here is the timeline at a glance, including the postponement introduced by the Digital Omnibus.
Since February 2025: Prohibited Practices and AI Literacy
Since 2 February 2025, the prohibited AI practices are enforceable. Anyone using or offering a banned application risks a sanction. At the same time, Article 4 of the AI Act already applies: the AI-literacy obligation.
The Digital Omnibus reworded Article 4. The original text required organisations to ensure, 'to their best extent', a sufficient level of AI literacy among their staff. It now says that providers and deployers of AI systems shall take measures to support the development of AI literacy of their staff and of other persons dealing with AI systems on their behalf. It adds explicitly that this does not require them to guarantee any specific level of AI literacy of any individual. In other words: an obligation to make an effort, not an obligation to achieve a result. The Commission and the member states are to support businesses, especially SMEs, in doing so.
Does that take the duty off the table? No. You still have to take measures, and that is exactly what a regulator will ask about: what have you done, for whom, and does it fit the AI you use? The law does not prescribe a certificate. An internal AI policy backed by training, plus a short record of who completed what, is the practical way to meet it.
Since 2 August 2026: Transparency, General-Purpose AI Models, and Oversight
Since 2 August 2026 a significant part of the AI Act has been genuinely enforceable. The Digital Omnibus did not postpone this. Since that date the transparency obligations under Article 50 apply: people must know when they are dealing with an AI system, and deepfakes must be disclosed as such.
At the same time, regulators gained teeth. The obligations for providers of general-purpose AI (GPAI) models, such as large language models, have applied since 2 August 2025. Since 2 August 2026 the European Commission can also enforce them and impose fines of up to 15 million euros or 3% of worldwide annual turnover. National market-surveillance authorities have been active since then too. For most SMEs this mainly means one thing: make sure your AI chatbot and any AI-generated content are clearly recognisable.
2 December 2026: Machine-Readable Marking and a New Ban
One part of Article 50 has a transition period: the duty to mark AI-generated content in a machine-readable way (Article 50(2)), for example through watermarks or metadata. That duty sits with the provider of the generative AI system, not with you as a user. Generative systems placed on the market on or after 2 August 2026 must comply from launch. Systems already on the market before 2 August 2026 have until 2 December 2026.
On the same date a new prohibition, added to Article 5 by the Digital Omnibus, takes effect: AI systems that generate child sexual abuse material or intimate images of people without their consent. For the typical SME user little changes, but it shows that the list of prohibited practices can grow.
Postponed to Late 2027 and 2028: The High-Risk Obligations
This is the biggest change introduced by the Digital Omnibus. The heavy obligations for high-risk AI systems were originally due to apply from 2 August 2026, but have been postponed. Standalone high-risk systems (Annex III: think of AI for recruitment screening, credit assessment, or education) now only need to meet the requirements from 2 December 2027.
For high-risk AI embedded in regulated products (Annex I, for example machinery or medical devices), the deadline moves to 2 August 2028. The postponement gives providers and regulators more time to finalise the accompanying standards and conformity assessments. Treat it as breathing room, not cancellation. It mainly affects providers. But if you use such a system yourself, for example a tool that ranks job applicants, you will have obligations of your own as a deployer from 2 December 2027, including organising human oversight, keeping logs and informing your staff in advance.
What Does the AI Act Mean in Practice for Your SME?
I often hear directors assume the AI Act only applies to tech companies or large platforms. That is a misconception. The AI Act applies to everyone who provides or uses AI systems in the EU. That includes an SME that uses ChatGPT, Copilot, or an AI recruitment tool.
The good news: most SMEs are deployers (users) of AI systems, not providers (developers). That means a lighter regime. And most tools you use day-to-day (text generators, AI-powered search, email assistants) fall into the limited or minimal risk categories. The heavy high-risk obligations rarely apply directly to most SMEs.
What does apply:
- Prohibited practices: check whether you use AI anywhere for social scoring, manipulation, emotion recognition of employees, or real-time biometric identification. Probably not, but be certain.
- AI literacy (Art. 4): take measures so that staff who work with AI understand the basics, and record what you have done. This has applied since February 2025. You do not have to guarantee a level per person, but you do have to make a real effort.
- Transparency (Art. 50, since August 2026): if you have an AI chatbot on your website, make clear that people are talking to AI. If you publish deepfakes, disclose that they were made with AI.
- High-risk check: do you use AI for recruitment screening, credit assessment, or medical support? Then stricter requirements apply from 2 December 2027, both for the provider (who carries out the conformity assessment) and for you as the deployer.
- Supplier check: if you procure high-risk AI, the provider must supply conformity documentation. Ask for it.
A Practical Action Plan to Become Compliant
You do not need to have everything in order all at once. Here is a workable sequence I recommend.
- Step 1: Take stock of your AI use. Which tools are in use, by whom, and for what purposes? This gives you an AI inventory that also serves as the foundation for your AI policy. How to organise that as a board is covered in AI governance for SMEs.
- Step 2: Determine the risk category for each application. Use the four tiers as your guide. Most tools are minimal or limited risk. Note the exceptions, and flag anything that talks to customers or creates content: the transparency rules have applied there since August 2026.
- Step 3: Address AI literacy. Article 4 has applied since February 2025 and requires you to take measures. Run an awareness session, put an internal AI policy in place, and designate a point of contact. A general session on what AI is rarely sticks. In an AI implementation I therefore train teams on the tools they use every day.
- Step 4: Check suppliers for high-risk AI. Ask your AI vendors whether their systems are classified as high-risk and what conformity documentation is available.
- Step 5: Draft an AI policy. Document which tools are approved, how personal data is handled, and how human oversight is organised. Align this with your GDPR policy.
- Step 6: Schedule an annual review. The AI Act evolves and implementing acts will follow. Set a recurring moment each year to test your compliance.
Common Misconceptions About the AI Act
Quite a few misunderstandings are circulating about the AI Act. The ones I encounter most often:
- "The AI Act only kicks in a few years from now." Incorrect: the prohibited practices and the AI-literacy obligation have applied since February 2025, and the transparency obligations and oversight of general-purpose AI models since 2 August 2026. Only the heavy high-risk obligations have been postponed via the Digital Omnibus, to December 2027 and August 2028.
- "The Digital Omnibus scrapped AI literacy." Incorrect: Article 4 was reworded, not removed. You still have to take measures. What has gone is any suggestion that you must guarantee a specific level for each employee.
- "We don't build AI ourselves, so it doesn't affect us." Incorrect: the law also applies to deployers: companies that use AI systems built by third parties.
- "Only tech companies need to do something." Incorrect: the AI Act cuts across every sector. An HR system, a credit check tool, or an AI chatbot on your website can already bring you into scope.
- "We'll wait for our supplier to sort it out." Partly true for conformity documentation. But the AI-literacy obligation and the ban on prohibited practices rest with you as the user.
- "Fines are only for large companies." The AI Act sets graduated fines based on the infringement, not exclusively on company size. That said, regulators are likely to enforce proportionately.
- "Using ChatGPT is high risk." No: a text generator is minimal or limited risk. High risk is context-dependent: it is about the purpose for which you deploy AI, not the tool itself.
Key takeaways
- The AI Act (EU 2024/1689) has been in force since August 2024 and is being phased in. The Digital Omnibus (Regulation (EU) 2026/1744, in force since 27 July 2026) moves the high-risk obligations to 2 December 2027 (Annex III) and 2 August 2028 (Annex I).
- Prohibited AI practices and the AI-literacy obligation (Art. 4) have applied since February 2025. From 2 December 2026 a ban on AI that generates child sexual abuse material or non-consensual intimate imagery is added.
- Since 2 August 2026 the transparency obligation (Art. 50) and enforcement of general-purpose AI (GPAI) models apply. That part was not postponed. Only machine-readable marking has until 2 December 2026, and only for generative systems already on the market.
- Most SMEs are deployers of limited or minimal risk AI, so the heavy high-risk obligations rarely apply to them directly.
- Article 4 is an obligation to make an effort: take measures so that staff who work with AI understand the basics. You do not have to guarantee a level per person, but a regulator will ask what you have done.
- Start with an AI inventory, determine the risk category of your applications, and put an internal AI policy in place.
Tackle Your AI Strategy and Governance Together?
In the boardroom workshop, we spend one day building a concrete AI strategy for your organisation, including the conditions, roles, and first steps for working AI governance aligned with the AI Act.
View the Boardroom Workshop