The AI Act is EU Regulation 2024/1689 on artificial intelligence. If you run a small business and use a chatbot or generate images, it applies to you. Start by working out which role you play.
Who supplies it, who uses it
A provider develops an AI system, or has one developed, and puts it on the market or into service under its own name or trade mark. A deployer uses a system under its own authority, for work. Buy a tool and use it as it comes, and you are a deployer.
Commission guidelines of 20 July 2026, which are not binding, mark out the edges. Build a chatbot in-house and run it under your own name, and you are a provider. The same goes if you modify someone else’s system and put it into service under your name.
What applies today
The bans in Article 5 have applied since 2 February 2025. They cover, among other things, harmful manipulation, exploiting vulnerabilities and social scoring. One matters to anyone with staff: no AI to infer people’s emotions at work, except for medical or safety reasons. Outside work and education this ban does not apply, but if a system recognises emotions or sorts people from biometric data, you have to tell them (Article 50, from 2 August 2026).
AI literacy, in Article 4, has also applied since 2 February 2025. You have to take steps that help the people using AI on your behalf to understand it. The Omnibus softened this: you need not guarantee any particular level.
Transparency, in Article 50, has applied since 2 August 2026. The notice has to be clear and distinguishable, and come at the first interaction or first viewing at the latest. The chatbot on your website has to say it is an AI, unless that is obvious. That duty is the provider’s. If you bought it, check that it does so. If it is your own, the job falls to you.
With images, audio and video, anyone using AI to generate or alter a deep fake has to say so. A deep fake is an image, audio or video made or altered by AI that resembles existing persons, objects, places, entities or events and would falsely appear genuine or true. The guidelines say an invented person can count, if plausible.
One of their examples is an advertising image that makes a product look better than it is. A notice does not make misleading advertising lawful. And if you show a real person, their rights over their image and voice still apply.
A real product on a generated background stays outside the rule, provided it does not mislead. Minor retouching such as colour or format usually stays outside too. If you hand advertising to an agency without deciding whether and how it uses AI, you are not the deployer; with a freelancer under your control, you still are. For evidently creative or satirical works the notice can be lighter, but advertising only qualifies in some cases.
Under the guidelines, a deep fake generated before 2 August 2026 need not be labelled after the fact, though the Commission encourages it.
For text, the duty applies only where it informs the public on matters of public interest and nobody has reviewed it and taken responsibility for it. Advertising and product descriptions usually stay outside, unless they make claims about health, consumer safety or sustainability.
What the Omnibus changed
The Digital Omnibus on AI is Regulation (EU) 2026/1744, published on 24 July 2026 and in force from the 27th.
It moved the rules on high-risk systems. For the uses in Annex III, such as filtering job applications, they start on 2 December 2027. For AI built into products such as lifts or toys, on 2 August 2028.
Providers of generative systems placed on the market before 2 August 2026 have until 2 December 2026 for machine-readable marking. The delay does not cover the notice you have to give. Chatbots and deep fakes got no delay. From 2 December 2026, AI systems that generate or alter intimate or sexual content of an identifiable person without their consent, or child sexual abuse material, are also banned.
Italy and smaller firms
Italian Law 132 of 23 September 2025 has been in force since 10 October 2025. Article 11(2) asks employers, or those commissioning work, to tell workers about their use of AI in the cases and ways set by Article 1-bis of Legislative Decree 152/1997, which covers fully automated systems used to decide or monitor in the employment relationship.
The law also created an offence, Article 612-quater of the Criminal Code. It covers unjust harm from images, video or voices falsified or altered with AI, capable of deceiving and spread without the consent of the person harmed.
Legislative Decree 179/2026, in force since 23 October 2026, aligns Italian rules, among other things on the powers of the authorities, AgID and ACN, and it also matters to anyone with staff. If you use AI in the employment relationship, decisions on starting, changing or ending employment, including disciplinary measures and performance assessments, cannot rest on automated processing alone: the final decision is for a person (Article 41(1)). Dismissals and other decisions taken in breach of this rule are void (Article 41(4)). A dismissal or other decision taken in breach of this rule is void. Searching for and screening job applications fall outside this rule, but not the others: the duty to inform, data protection and, from 2 December 2027, the high-risk rules.
If AI affects how work is organised or its pace, it has to be covered in the risk assessment under Article 28 of Legislative Decree 81/2008 (Article 42(2)). For other Italian rules and for penalties, ask your lawyer.
For SMEs, the maximum fine is the lower of the percentage of worldwide annual turnover and the fixed amount (Article 99): for Article 50, 3% or €15 million; for the Article 5 bans, 7% or €35 million. The Regulation asks Member States for measures for smaller firms: training, channels for questions, priority access to regulatory sandboxes. The Commission must publish practical examples for Article 4. Article 50, though, does not exempt small firms.
What to do now
This is advice, not law.
- List your AI tools and note which ones you have modified or run under your own name.
- Explain to the people using them what they do and where they go wrong, and make a note of when you did.
- Check that your chatbot clearly says it is an AI in its first message. The word “assistant” alone is not enough, and neither is a line in your terms.
- Settle on a plain sentence or label for realistic images, video and voices made with AI, and put it on the content, where it can be seen or heard. A line in your terms or the tool’s invisible mark is not enough.
- If you use AI to select or assess people, talk to your employment adviser.
This is a general overview, not legal advice. Bringing AI into businesses that make things is our work and these rules are part of it, but on your own case, check with your lawyer.
Sources
- Regulation (EU) 2024/1689 on artificial intelligence, Official Journal of the European Union, 12 July 2024. eur-lex.europa.eu
- Regulation (EU) 2026/1744, Digital Omnibus on AI, Official Journal of the European Union, 24 July 2026. eur-lex.europa.eu
- Guidelines on the implementation of the transparency obligations for certain AI systems under Article 50 of Regulation (EU) 2024/1689, annex to C(2026) 5054 final, European Commission, 20 July 2026. ai-act-service-desk.ec.europa.eu
- AI Literacy, Questions & Answers, European Commission, updated 20 October 2026. digital-strategy.ec.europa.eu
- Navigating the AI Act, European Commission, updated 20 October 2026. digital-strategy.ec.europa.eu
- AI Act, Shaping Europe’s digital future, European Commission, updated 19 October 2026. digital-strategy.ec.europa.eu
- Law of 23 September 2025, no. 132, Disposizioni e deleghe al Governo in materia di intelligenza artificiale, Normattiva, Gazzetta Ufficiale no. 223 of 25 September 2025. www.normattiva.it
- Legislative Decree of 7 October 2026, no. 179, Normattiva, Gazzetta Ufficiale no. 234 of 8 October 2026. www.normattiva.it
- EC News editorial staff, Euroconference, “Utilizzo di sistemi di AI in azienda: il D.Lgs. no. 179/2026 introduce nuovi obblighi”, 13 October 2026. www.ecnews.it
- Luca De Menech, Martino Ruggiero, Dentons, “Intelligenza Artificiale e diritto del lavoro. Il D. Lgs. no. 179/2026 tra tutele dei lavoratori, sicurezza sul lavoro e prospettive”, dirittobancario.it, 27 October 2026. www.dirittobancario.it
- Irene Salvi, “Legge italiana sull’intelligenza artificiale: in vigore il secondo decreto attuativo”, ICT Security Magazine, 29 October 2026. www.ictsecuritymagazine.com
- Stefano De Luca Tamajo, Toffoletto De Luca Tamajo, Ius Laboris, “Italy’s First Law On Artificial Intelligence Takes Effect”, Legal 500, 25 November 2025. www.legal500.com
- Emanuele Martinoli, “L’impatto dell’intelligenza artificiale sugli obblighi informativi del datore di lavoro”, EC News, Euroconference, 1 December 2025. www.ecnews.it
- Decree-Law of 4 May 2023, no. 48, Article 26, amending Article 1-bis of Legislative Decree 152/1997, text on Brocardi.it. www.brocardi.it
- Codice penale (Italian criminal code), Article 612-quater, Illecita diffusione di contenuti generati o alterati con sistemi di intelligenza artificiale, text on Brocardi.it. www.brocardi.it