EU AI Act explained: risk levels, deadlines, fines and what your company must do
A plain-language overview of Regulation (EU) 2024/1689, the EU Artificial Intelligence Act, for companies that build, sell or use AI. Updated November 2026, including the changes made by Regulation (EU) 2026/1744 (the “Digital Omnibus on AI”). Every date and number below points to the article it comes from.
- What it is: the first comprehensive AI law in the world. It regulates AI by risk: some uses are banned, “high-risk” uses get strict duties, and AI that talks to people or creates content must be transparent.
- Who it covers: anyone who places AI on the EU market or uses it in the EU, and companies outside the EU whose AI output is used in the EU (Art. 2).
- Already in force: the bans and AI literacy (since 2 February 2025), rules for general-purpose AI models (since 2 August 2025) and the transparency rules of Article 50 (since 2 August 2026).
- Still coming: high-risk rules for Annex III uses on 2 December 2027 and for AI in regulated products on 2 August 2028.
- Fines: up to EUR 35 million or 7 % of worldwide annual turnover for banned practices (Art. 99).
1. The four risk levels
The AI Act does not regulate “AI” as a whole. It regulates what an AI system is used for. The same model can be harmless in one product and high-risk in another.
| Risk level | Examples | What the law requires |
|---|---|---|
| Unacceptable | Social scoring, manipulative techniques that cause harm, emotion recognition at work or school, untargeted scraping of faces | Banned (Art. 5) |
| High | AI in recruitment, credit scoring, education, critical infrastructure, law enforcement, migration, justice; AI as a safety component of regulated products | Risk management, data governance, technical documentation, logging, human oversight, accuracy and robustness, conformity assessment, registration (Art. 6–49) |
| Transparency | Chatbots, generators of images, audio, video and text, deep fakes, emotion recognition | Tell people they deal with AI; mark AI-generated content in a machine-readable way; disclose deep fakes (Art. 50) |
| Minimal | Spam filters, AI in video games, most internal tools | No new obligations beyond AI literacy (Art. 4) |
On top of the four levels, providers of general-purpose AI models (the large models behind chatbots and generators) have their own duties: technical documentation, a copyright policy and a summary of training content, and for the most capable models also risk assessment and incident reporting (Art. 53–55).
2. The timeline
| Date | What applies | Source |
|---|---|---|
| 1 August 2024 | The AI Act enters into force. | Art. 113 |
| 2 February 2025 | Bans on prohibited practices; AI literacy. | Art. 113(a) |
| 2 August 2025 | Rules for general-purpose AI models; governance; penalties. | Art. 113(b) |
| 2 August 2026 | The Act applies in general, including the transparency duties of Article 50. | Art. 113 |
| 2 December 2026 | Generative AI systems placed on the market before 2 August 2026 must mark their outputs (Art. 50(2)); two new bans on AI-generated non-consensual intimate images and child sexual abuse material apply. | Reg. (EU) 2026/1744 |
| 2 August 2027 | General-purpose AI models placed on the market before 2 August 2025 must comply. | Art. 111(3) |
| 2 December 2027 | High-risk rules for the uses listed in Annex III (recruitment, credit, education, biometrics and others). | Reg. (EU) 2026/1744 |
| 2 August 2028 | High-risk rules for AI that is a safety component of products covered by Annex I (machinery, toys, medical devices and others). | Reg. (EU) 2026/1744 |
The Digital Omnibus moved the high-risk dates; it did not remove the obligations. Companies that start late will have the same work in less time.
3. What is banned
Since 2 February 2025 these practices are prohibited in the EU (Art. 5):
- subliminal, manipulative or deceptive techniques that distort behaviour and cause significant harm;
- exploiting the vulnerabilities of people because of age, disability or social or economic situation;
- social scoring that leads to unjustified or disproportionate treatment;
- predicting that a person will commit a crime based solely on profiling or personality traits;
- building facial recognition databases by untargeted scraping of images from the internet or CCTV;
- emotion recognition in the workplace and in education, except for medical or safety reasons;
- biometric categorisation that infers race, political opinions, trade union membership, religious beliefs, sex life or sexual orientation;
- real-time remote biometric identification in public spaces for law enforcement, with narrow exceptions.
From 2 December 2026 the Digital Omnibus adds AI systems that generate non-consensual intimate imagery (“nudifier” apps) and child sexual abuse material.
4. Who has which duties
| You are | Meaning | Main duties |
|---|---|---|
| Provider | You develop an AI system or model, or have it developed, and place it on the market under your name. | Most of the obligations: conformity for high-risk systems, marking of generated content, documentation. |
| Deployer | You use an AI system in your professional activity. | Use high-risk systems according to instructions, with human oversight; disclose deep fakes and AI-written public-interest texts; AI literacy of your staff. |
| Importer / distributor | You bring a non-EU AI system to the EU market or resell it. | Check that the provider met its obligations before you sell. |
Most companies are deployers: they use AI tools made by others. A company that puts its own brand on an AI product, or substantially changes one, can become a provider (Art. 25).
5. Fines
| Breach | Maximum fine | Source |
|---|---|---|
| Banned practices | EUR 35 000 000 or 7 % of worldwide annual turnover | Art. 99(3) |
| Most other obligations, including Article 50 transparency | EUR 15 000 000 or 3 % | Art. 99(4) |
| Incorrect or misleading information to authorities | EUR 7 500 000 or 1 % | Art. 99(5) |
| Providers of general-purpose AI models | EUR 15 000 000 or 3 % | Art. 101 |
For large companies the higher of the two amounts applies; for SMEs and start-ups the lower one (Art. 99(6)). The Digital Omnibus extends SME-style relief to “small mid-caps”.
6. A checklist for your company
- Inventory. List every AI system you build, buy or use, and what it is used for.
- Classify. For each one decide the risk level and whether you are provider, deployer or both.
- Stop anything banned. This has been mandatory since February 2025.
- AI literacy. Take measures so that the people who operate AI understand it (Art. 4).
- Transparency now. Since August 2026: tell users when they talk to a chatbot, mark AI-generated content in a machine-readable way and disclose deep fakes. Our Article 50 guide explains it step by step.
- High-risk later, but start now. If you use AI in hiring, credit, education or another Annex III area, plan documentation, oversight and testing before December 2027.
- Keep evidence. Authorities ask what you did and when. Records made at the time count more than records made after the fact.
7. Where Ankorith fits
Ankorith covers the part of the AI Act that already applies to almost every company that publishes content: Article 50. It writes a machine-readable “AI-generated” label into the file in the IPTC format platforms read, keeps a fingerprint so the file is recognised even after social networks strip the metadata, shows a verified company as the issuer, and records every seal in a public, tamper-evident log as evidence of what you labelled and when.
Try Ankorith: verification is free
This page is an overview, not legal advice. How the AI Act applies to a specific product depends on its intended purpose; if in doubt, ask your lawyer.
Frequently asked questions
Does the EU AI Act apply to companies outside the EU?
Yes, if they place AI systems on the EU market or if the output of their AI is used in the EU (Art. 2(1)).
Was the AI Act postponed?
Only partly. The Digital Omnibus (Regulation (EU) 2026/1744) moved the high-risk rules to 2 December 2027 and 2 August 2028. The bans, the rules for general-purpose AI models and the transparency rules of Article 50 were not postponed.
Do I have to label AI-generated images and videos?
Providers of generative AI must mark outputs in a machine-readable way, and companies that publish deep fakes must disclose them (Art. 50(2) and 50(4)). See the Article 50 guide.
Is ChatGPT high-risk under the AI Act?
A general-purpose model is not high-risk by itself; its provider has the duties for general-purpose AI models. A system built on it becomes high-risk when it is used for a purpose listed in the Act, such as screening job applicants.
Sources
- Regulation (EU) 2024/1689 (Artificial Intelligence Act), in particular Articles 2, 4, 5, 6, 25, 50, 53–55, 99, 101, 111 and 113 and Annexes I and III — EUR-Lex.
- Regulation (EU) 2026/1744 (Digital Omnibus on AI), published in the Official Journal on 24 July 2026, in force since 27 July 2026.