This is a practical guide, not legal advice. For decisions about your specific case, talk to a lawyer. Sources last checked on 2 November 2026.
The EU AI Act (Regulation (EU) 2024/1689) sets rules for AI according to risk. For a typical SMB that uses ChatGPT, Microsoft Copilot or AI agents, it means four things today: support your staff's AI literacy, stay away from the banned practices, tell people when they deal with AI or AI-generated content, and check whether any use falls in the high-risk list, which now applies from 2 December 2027.
For everyday tools the obligations are light: most of the work is knowing what you use, who uses it and with which data, and writing it down. Below, each point separates what the law says, what official guidance says and what we recommend at LeaderNova.
Does the AI Act apply to a small company?
Yes, if you use AI in your business. The Act was published in the Official Journal on 12 July 2024 and entered into force on 1 August 2024 (EUR-Lex). It applies by role, whatever the company size. A provider develops an AI system and places it on the market or puts it into service under its own name: OpenAI, Microsoft, Anthropic and Google for their tools. A deployer uses an AI system under its own authority in a professional activity: a company that gives its team ChatGPT or Copilot.
One catch: if you build a customer chatbot on top of a model API and put it into service under your name, you may be its provider, with the provider obligations that come with it.
Which dates apply, after the Digital Omnibus?
The AI Act applies in stages. In July 2026 the EU amended the timeline with the Digital Omnibus on AI, Regulation (EU) 2026/1744 of 8 July 2026, published in the Official Journal on 24 July 2026 (EUR-Lex). It entered into force on the third day after publication, 27 July 2026. The main change for companies: the high-risk rules moved back.
| Date | What applies |
|---|---|
| 2 February 2025 | Prohibited practices (art. 5) and AI literacy (art. 4) |
| 2 August 2025 | Rules for general-purpose AI model providers, governance, national penalty regimes |
| 2 August 2026 | Transparency obligations (art. 50) |
| 2 December 2026 | Machine-readable marking (art. 50(2)) for generative systems already on the market before 2 August 2026; two new prohibitions (AI-generated non-consensual intimate imagery and child sexual abuse material) |
| 2 December 2027 | High-risk systems in Annex III: employment, education, credit and others (originally 2 August 2026) |
| 2 August 2028 | High-risk AI in regulated products, Annex I (originally 2 August 2027) |
The high-risk dates are in recital 40 of Regulation 2026/1744, the art. 50(2) transition in recital 38. The December 2026 date for the new prohibitions comes from law-firm summaries (Orrick, July 2026); we could not confirm it in the operative text on EUR-Lex on 2 November 2026.
The four risk levels in plain terms
Prohibited. Banned since February 2025: harmful manipulation, exploiting vulnerabilities such as age or disability, social scoring, emotion recognition at work or in schools (except for medical or safety reasons), untargeted scraping of facial images, and a few others. Few SMBs do any of this, but check HR and call tools: "mood detection" features on staff calls can fall in the emotion recognition ban.
High-risk. AI used for decisions that seriously affect people. For SMBs the categories to watch in Annex III are recruitment and worker management (screening CVs, evaluating candidates, deciding promotions or terminations, monitoring performance), creditworthiness assessment, and risk assessment and pricing in life and health insurance. From 2 December 2027 these uses bring deployer obligations under art. 26: follow the provider's instructions, assign human oversight to competent people, keep the logs, and inform workers and their representatives before using the system at work. If you already use AI this way, plan for it now and ask the provider how they will meet the Annex III requirements.
Transparency. AI that talks to people or produces content people may take as real: chatbots, synthetic images, audio and video, deepfakes. Covered by art. 50 from 2 August 2026 (details below).
Minimal risk. Everything else, which is most of what an SMB does with AI: drafting emails, summarising documents, code assistants, internal search, translation, spam filters. The AI Act adds no specific obligations here beyond AI literacy. GDPR still applies whenever personal data goes into these tools.
The models behind ChatGPT, Copilot, Gemini or Claude have had their own rules since August 2025; those fall on the model providers, and using the models through apps or APIs doesn't make you one.
What a typical SMB using ChatGPT, Copilot or agents has to do
1. Support AI literacy
What the law says: art. 4, as amended by Regulation 2026/1744, requires providers and deployers to "take measures to support the development of AI literacy" of their staff and of others who operate or use AI systems on their behalf, considering their technical knowledge, experience, training and the context of use. The amended text states that this "does not require providers or deployers to guarantee any specific level of AI literacy of any individual." It has applied since 2 February 2025. Before the amendment, the article asked companies to take measures to ensure, to their best extent, a sufficient level of AI literacy; the amended text asks them to support it.
What guidance says: the Commission's AI literacy Q&A (published 7 May 2025, before the amendment) says there is no need for a certificate and that organisations can keep an internal record of training and other initiatives. It suggests covering what AI is, the company's role and the risks of the systems used.
Our recommendation: a short session per role, built around the tools people actually use and the data rules that apply to them. Keep a simple record: who attended, what was covered, when. Repeat it when tools change.
2. Keep an AI-use inventory
What the law says: the AI Act has no standalone register requirement for minimal-risk use. You still need to know your role and risk category for each use, and under GDPR art. 30 your records of processing should include processing that runs through AI tools.
Our recommendation: a one-page register with one row per tool: tool and plan (free, business, enterprise, API), the owner who answers for it, what it's used for, which data is allowed, and the risk level. It usually turns up tools nobody approved.
3. Be transparent where art. 50 applies
What the law says: from 2 August 2026, art. 50 requires that:
- people are informed when they interact with an AI system, unless it's obvious (an obligation on the provider of the system: if you built the chatbot, that's you);
- providers of generative AI mark synthetic audio, image, video and text in a machine-readable way;
- deployers of emotion recognition or biometric categorisation inform the people exposed to it;
- deployers who publish deepfakes disclose that the content is artificially generated or manipulated;
- deployers who publish AI-generated text to inform the public on matters of public interest disclose it, unless the text went through human review or editorial control and someone holds editorial responsibility.
What guidance says: the Commission's Guidelines on transparency obligations (20 July 2026) and a Code of Practice on marking and labelling AI-generated content explain how to apply these rules.
Our recommendation: if a customer can chat with an AI on your site, by email or on WhatsApp, say so at the start of the conversation, in the conversation itself. Label AI-generated images and video of real-looking people. Ordinary marketing copy that a person reviewed and signed off is usually outside the public-interest text rule; news-style content and public statements deserve a closer look.
4. Check your vendors, including the GDPR contract
What the law says: when an AI tool processes personal data on your behalf, GDPR art. 28 requires a processor that gives sufficient guarantees and a contract, the data processing agreement (DPA). Transfers outside the EU need a legal basis under GDPR chapter V.
Our recommendation: check the plan as well as the brand. Consumer plans of AI tools usually run on consumer terms, often without a DPA; business, enterprise and API plans usually offer one. For each tool check the DPA, whether your inputs train the model (and how to turn it off), where data is stored, the sub-processors and retention. Until a tool passes, no client or employee data goes into it.
5. Write a short AI policy
What the law says: the AI Act doesn't require a written AI policy for minimal-risk use. In Italy, Law 132/2025 (below) requires employers to inform workers about the use of AI, and professionals to inform clients about the AI systems they use.
Our recommendation: two pages are enough. Approved tools and plans; what data can go where; which outputs stay drafts until a person sends them; who approves agents that act on their own; which actions are logged. For AI agents add a written fence: what each agent can read, what it can change, and what needs a human approval. Unlimited access is a named risk in the OWASP Top 10 for LLM Applications (LLM06, Excessive Agency).
What Italy adds: Law 132/2025
Italy passed its national AI law, Legge 23 settembre 2025, n. 132 ("Disposizioni e deleghe al Governo in materia di intelligenza artificiale"), published in the Gazzetta Ufficiale n. 223 of 25 September 2025 and in force since 10 October 2025 (Normattiva). Article 1 says it must be applied consistently with the AI Act. For SMBs the relevant points are:
- Art. 11: the employer must inform workers about the use of AI, in the cases and ways set by Italian worker-information rules.
- Art. 13: professionals in the intellectual professions (lawyers, accountants, engineers and others) must tell clients, clearly and completely, which AI systems they use.
- Art. 20: AgID and ACN are the national authorities for AI.
What are the penalties?
Under art. 99, fines reach €35 million or 7% of worldwide annual turnover for prohibited practices, and €15 million or 3% for most other obligations, including art. 50. For SMEs the cap is the lower of the two amounts; the Omnibus extends tailored treatment to small mid-caps. Art. 99 sets no specific fine for AI literacy. According to the Commission's AI literacy Q&A, penalties for art. 4 come from national law, and national market surveillance authorities supervise and enforce the rules from 2 August 2026.
AI Act checklist for SMBs
| Action | Legal basis | Evidence to keep | Applies from |
|---|---|---|---|
| AI literacy measures for staff and contractors who use AI | AI Act art. 4 | Training record: who, what, when | 2 Feb 2025 |
| Check no tool does emotion recognition at work or other banned practices | AI Act art. 5 | Note in the AI-use register | 2 Feb 2025 |
| AI-use register: tool, plan, owner, purpose, data, risk level | Our recommendation; supports GDPR art. 30 | The register, reviewed quarterly | Now |
| DPA and data checks for every AI tool that touches personal data | GDPR art. 28, chapter V | Signed DPA, settings screenshots, sub-processor list | Now |
| Disclose AI chatbots at the start of the conversation | AI Act art. 50(1) | Screenshot of the disclosure | 2 Aug 2026 |
| Label deepfakes and public-interest AI text | AI Act art. 50(4) | Labelling rule in the AI policy | 2 Aug 2026 |
| Inform workers about AI use (Italy) | Law 132/2025 art. 11 | Written notice to staff | 10 Oct 2025 |
| Inform clients about AI use (Italian professionals) | Law 132/2025 art. 13 | Clause in engagement letters | 10 Oct 2025 |
| Written AI policy with approvals and agent limits | Our recommendation | The policy, version and date | Now |
| Plan for high-risk uses (hiring, staff evaluation, credit) | AI Act art. 26, Annex III | Provider answers, oversight plan | 2 Dec 2027 |
Where we can help
We write AI policies under art. 4 and run vendor and DPA checks as part of our security and compliance work. Our AI business automation service builds workflows where every output stays a draft until a person approves it. To see where your company stands, book a 30-minute strategy call: we'll go through the tools you use and tell you what is missing. Book a strategy call
Sources
- Regulation (EU) 2024/1689 (AI Act), OJ 12 July 2024: EUR-Lex
- Regulation (EU) 2026/1744 (Digital Omnibus on AI), 8 July 2026, OJ 24 July 2026: EUR-Lex
- European Commission, AI literacy Q&A, 7 May 2025: digital-strategy.ec.europa.eu
- European Commission, Guidelines on transparency obligations for providers and deployers of AI systems, 20 July 2026: digital-strategy.ec.europa.eu
- Legge 23 settembre 2025, n. 132, GU n. 223 of 25 September 2025: Normattiva
- Orrick, "EU AI Act Update: Digital Omnibus Finalizes 8 Compliance Changes", July 2026: orrick.com
Last checked: 2 November 2026.