What does the AI Act change for me, and when?
The EU AI Act after the 2026 Digital Omnibus: what applies today, what moved to 2027 and 2028, and what Italy's AI law adds. Dates from the official texts.
AI-generated image: Internet of Things News / OpenAI ChatGPT
After reading this article you will be able to:
- Say which AI Act rules already apply to you, and since which date
- Tell which rules the Digital Omnibus postponed, and to when
- Know what Italy's AI law adds and which authority handles what in Italy
Drafted with AI from primary sources · checked by a separate AI agent · spot-checked by our editors · How we check
On this page
The AI Act, the European Union’s law on artificial intelligence, already changes a few things you can notice, and the bigger changes come later. Since 2 August 2026, a chatbot must tell you that it is a machine, and fake photos, videos and voices made with AI for work or business must be labelled. A short list of harmful uses has been banned since 2 February 2025, and from 2 December 2026 the ban also covers apps that make fake nude images of real people. The strict rules for AI that helps decide who gets a job, a loan or a place at school were moved to 2 December 2027. In Italy, a national law has added its own rules since 10 October 2025, including prison for spreading harmful AI fakes.
Every date here comes from the official texts, on EUR-Lex, in Italy’s Gazzetta Ufficiale and on European Commission pages, opened on 15 November 2026. This page explains the rules; it is not legal advice.
This site is written with AI tools, including Claude by Anthropic. This page uses only what makers, public authorities and published studies say, with links.
What is the AI Act, and why did its dates move?
The AI Act is Regulation (EU) 2024/1689. It entered into force on 1 August 2024, and its rules start in stages. It sorts AI by the harm it can do: a few uses are banned, uses that can change someone’s life get strict rules, and some everyday tools, such as chatbots, only have to be honest about being AI.
In July 2026 the EU changed the timetable. The Digital Omnibus on AI, Regulation (EU) 2026/1744 of 8 July 2026, says the technical standards companies need were late and national authorities were not ready, so it moved the strict rules back by a year or more. It also added a new ban.
What already applies today?
Banned uses, since 2 February 2025 (Article 5). No one may sell or use AI that:
- uses hidden tricks to push people into decisions that seriously harm them, or exploits their age, disability or money problems to do so;
- gives people a “social score” that leads to unfair treatment;
- predicts whether someone will commit a crime only from their profile or personality;
- builds face databases from photos collected at random online or from security cameras;
- reads people’s emotions at work or at school, except for medical or safety reasons;
- guesses someone’s race, politics, religion or sexual orientation from their face or body.
Live face recognition by the police in public places is also banned, apart from a few serious cases, such as searching for a kidnapping victim, and normally only with the prior approval of a judge or an independent authority.
Training at work, since 2 February 2025 (Article 4). Organizations that use AI must help their staff understand it. What rules apply when a small business uses AI? explains what this means, with a training log to copy.
Rules for the big models, since 2 August 2025 (Article 53). The companies behind the large general-purpose models that power assistants such as ChatGPT, Claude or Gemini must keep technical documents, have a policy to respect copyright and publish a summary of what the model was trained on.
Honesty about AI, since 2 August 2026 (Article 50):
- A chatbot must tell you it is an AI, unless that is obvious.
- Anyone who uses AI for work or business to make a deepfake, a fake image, sound or video that looks real, must say it is artificial. Satire and fiction need only a light mention.
- AI-written texts published to inform the public on matters of public interest must be labelled, unless a person checked them and an editor takes responsibility.
- If a system reads your emotions or sorts you by your face or body, you must be told.
- Companies whose AI makes images, sound, video or text must add a hidden mark that computers can read (tools already on sale before 2 August 2026 have until 2 December 2026).
You must be told clearly, at the latest the first time you meet the AI or its content. The Commission’s questions and answers on Article 50, updated on 24 July 2026, add that content made before 2 August 2026 does not have to be labelled after the fact. How can you tell if a photo or video was made with AI? explains what labels and marks can show.
The right to complain, since 2 August 2026 (Article 85). Anyone who thinks the law was broken can complain to the national market surveillance authority, the public body that checks products and services on sale.
What was postponed, and what comes next?
| Date | What starts |
|---|---|
| 2 December 2026 | New ban, added by the Omnibus: AI that makes realistic sexual images, videos or sounds of a real person without their explicit consent (“nudify” apps), or child sexual abuse material. Also the deadline for AI tools on sale before 2 August 2026 to add hidden marks |
| 2 December 2027 (was 2 August 2026) | Strict rules for the high-risk uses in Annex III, including hiring and managing workers, school admissions and grading, credit scores, life and health insurance prices and public benefits |
| 2 August 2028 (was 2 August 2027) | Strict rules for AI inside products that already need EU safety checks, such as toys, lifts and medical devices |
The new ban covers the companies that sell such tools and anyone who uses them that way for work or business; purely private use is outside the AI Act, though spreading such images can be a crime, as in Italy. When the high-risk rules start, an organization that uses that kind of AI to decide, or help decide, about you must tell you, and must put a trained person in charge of watching over it (Article 26).
What does Italy add?
Law 132 of 23 September 2025, in force since 10 October 2025, adds:
- Children (Article 4). Under 14, a parent must consent to the child using AI. From 14 to 17, teenagers can agree themselves to the use of their data, if the information is easy to understand.
- Health (Article 7). AI supports doctors, but the decision “è sempre rimessa agli esercenti la professione medica” (“is always left to the medical profession”), and patients have the right to be told when AI is used.
- Work (Article 11). Employers must tell workers when they use AI in fully automated systems that help decide on hiring, tasks, evaluation or dismissal, or that monitor them.
- Professionals (Article 13). Lawyers, accountants and other professionals must tell clients which AI they use, in clear and simple language.
- Public offices and courts (Articles 14 and 15). A person stays responsible for every decision of a public office, and a judge for every ruling.
- Deepfakes (Article 26). A new crime: causing unjust harm to someone by spreading, without their consent, images, videos or voices faked with AI that can fool people. The penalty is one to five years in prison.
Two government decrees followed. Decree 160 of 2026, in force since 30 September 2026, helps people harmed by AI who sue for damages: if their claim is plausible, a judge can order the other side to show the AI’s records, and when the harm comes from breaking an AI Act duty, the link between the two is presumed unless the other side proves otherwise. Decree 179 of 2026, in force since 23 October 2026, sets the authorities’ powers and the fines.
Who enforces the rules in Italy?
- ACN, the National Cybersecurity Agency, checks AI on the market, can inspect and fine, and receives complaints.
- AgID, the Agency for Digital Italy, supervises the bodies that test high-risk AI.
- Banca d’Italia (the Bank of Italy), CONSOB (the stock market authority) and IVASS (the insurance authority) check high-risk AI used for loans, investments and insurance.
- The Garante, the data protection authority, keeps its powers over personal data and also checks some high-risk AI used by the police, at borders, in courts and in elections. Its AI page collects its decisions and guidance on AI.
A complaint sent to the wrong authority is passed on to the right one, and you are told. For a crime such as a harmful deepfake, file a complaint at a police station or with the carabinieri.
What applies to me today?
What applies on 25 November 2026. “EU” rules apply in every EU country, “Italy” rules only in Italy.
| If you are | What applies | Since |
|---|---|---|
| Anyone | Chatbots must say they are AI; deepfakes made for work or business must be labelled (EU) | 2 August 2026 |
| Anyone | You can complain about a breach, to ACN in Italy (EU) | 2 August 2026 |
| Anyone | Spreading a harmful AI fake of someone is a crime (Italy) | 10 October 2025 |
| Patient | You must be told when AI is used, and the doctor decides (Italy) | 10 October 2025 |
| Worker | No AI reading your emotions at work, except for medical or safety reasons (EU) | 2 February 2025 |
| Worker | Your employer must tell you when AI is used in automated systems that decide or monitor (Italy) | 10 October 2025 |
| Parent | Children under 14 need your consent to use AI (Italy) | 10 October 2025 |
| Parent | No emotion reading at school, and no AI exploiting a child’s age to harm them (EU) | 2 February 2025 |
| Small business | Help staff understand the AI they use (EU) | 2 February 2025 |
| Small business | Label deepfakes you publish; customer chatbots must say they are AI (EU) | 2 August 2026 |
| Professional | Tell clients which AI you use (Italy) | 10 October 2025 |
Not yet in force: the ban on “nudify” apps (2 December 2026) and the strict rules for AI that screens job applications, grades students or checks credit (2 December 2027).
Where does this answer stop?
- What the law leaves out. The AI Act does not cover AI used only for military, defence or national security, research before a product goes on sale, or your own personal, non-professional use (Article 2).
- Guidelines still to come. The Commission’s draft guidelines on which AI counts as high-risk came out on 19 May 2026; its consultation page says the final version will be adopted by the end of 2026. Until then, whether a tool is high-risk can be unclear.
- Labels in practice. The Commission’s guidelines on Article 50 (published on 20 July 2026, formally adopted on 5 November 2026) and a voluntary code of practice on labelling AI content (10 June 2026) explain how to comply. How strictly authorities will check is not yet known.
- Older systems. High-risk AI already on sale or in use before its start date has to follow the strict rules only if its design changes a lot; when public offices use it, it must comply by 2 August 2030 in any case (Article 111, as changed by the Omnibus).
- Compensation. The AI Act sets duties and fines; it does not pay anyone back. Whether you can claim damages is a question for a lawyer.
- Outside the EU the rules are different.
Where to go next
- EU AI Act: the law in a few lines, with links to both texts.
- What rules apply when a small business uses AI?: staff training, privacy and professional secrecy.
- What happens to what you type into ChatGPT, Gemini or Claude?: your data and the settings that control it.
- AI for everyone: the complete guide to today’s AI.
Common questions
When does the EU AI Act come into force?
It entered into force on 1 August 2024, but its rules start in stages. The bans and the duty to train staff apply from 2 February 2025, the rules for the large models behind chatbots from 2 August 2025, and most other rules, including telling people they are talking to an AI, from 2 August 2026. The strict rules for high-risk AI start on 2 December 2027 and 2 August 2028.
Has the AI Act been delayed?
Partly. The Digital Omnibus on AI, Regulation (EU) 2026/1744, moved the strict rules for high-risk AI from 2 August 2026 to 2 December 2027 for uses such as hiring, credit and school, and from 2 August 2027 to 2 August 2028 for AI built into products such as toys or medical devices. The bans and the transparency rules were not postponed. Only AI tools already on sale before 2 August 2026 got until 2 December 2026 to add hidden marks to what they generate.
Does the AI Act ban ChatGPT?
No. Chatbots are allowed in the EU. Since 2 August 2025 the companies that make the large models behind them have duties, such as a policy to respect copyright and a public summary of the content used to train the model. Since 2 August 2026 a chatbot must make clear that it is an AI. What the law bans is a short list of uses, such as manipulating people or giving them a social score.
Do I have to label AI images I post on social media?
The AI Act's labelling duty is for people and organizations that use AI for work or business; it does not cover a purely personal, non-professional use. Social networks can have their own labelling rules, so check theirs. In Italy, publishing or spreading an image, video or voice of a real person faked with AI, without their consent is a crime when it causes them unjust harm.
Is making a deepfake a crime in Italy?
Spreading one can be. Since 10 October 2025, the Italian criminal code punishes with one to five years in prison anyone who causes unjust harm to a person by giving, publishing or spreading, without their consent, images, videos or voices faked with AI that can fool people. As a rule, the victim must file a formal complaint (querela). For your own case, ask a lawyer or the police.
Topics: Who answers?, AI Act, AI rules, Italy, deepfakes