EU AI Act
Regulation (EU) 2024/1689, the European Union’s horizontal law on artificial intelligence, amended
in 2026 by the Digital Omnibus on AI, Regulation (EU) 2026/1744. It regulates AI systems by risk
level and by the role an organisation plays. This note covers how the Act is built. Dates, the
Omnibus amendments and other countries’ regimes are in ai-regulation-and-policy. Voluntary
frameworks that help implement it are IEC 42001 and
NIST AI RMF.
Risk levels
| Level | Treatment | Examples |
|---|---|---|
| Unacceptable | Prohibited (Art. 5) | Social scoring, manipulative AI, emotion recognition at work, untargeted face scraping; from 2 December 2026 also AI generating non-consensual sexual deepfakes or child sexual abuse material |
| High risk | Regulated (Chapter III; classification in Art. 6) | Annex III uses such as employment, education, credit scoring, critical infrastructure, biometrics; Annex I products such as machinery and medical devices |
| Limited risk | Transparency duties (Art. 50) | Chatbots must disclose they are AI; AI-generated or manipulated content must be marked |
| Minimal risk | No specific duties | Spam filters, most internal productivity uses |
General-purpose AI (GPAI) models sit beside this scale with their own chapter.
Roles
- Provider: develops an AI system or GPAI model and places it on the EU market or puts it into
service under its own name. Carries most obligations. Providers outside the EU are covered when
the system’s output is used in the EU. - Deployer: uses an AI system in a professional capacity. Lighter duties than a provider.
- Importer and distributor: bring or pass on systems in the EU supply chain.
A deployer that puts its name on a high-risk system, or substantially modifies one, can become its
provider.
High-risk obligations
Provider requirements (Arts. 8–17):
| Article | Requirement |
|---|---|
| 9 | Risk management system across the lifecycle |
| 10 | Data and data governance for training, validation and test data |
| 11 | Technical documentation |
| 12 | Record-keeping (automatic logs) |
| 13 | Transparency and instructions for use for deployers |
| 14 | Human oversight by design |
| 15 | Accuracy, robustness and cybersecurity |
| 17 | Quality management system |
Then conformity assessment, the EU declaration of conformity, CE marking and registration in the EU
database.
Deployer duties (Art. 26): use the system according to the instructions; assign human oversight to
people with the competence, training and authority for it; make sure input data under their control
is relevant and representative; monitor operation, and report risks and serious incidents; keep the
automatic logs for at least six months; inform workers and their representatives before use at the
workplace; inform people when an Annex III system makes or assists decisions about them; use the
provider’s information for a GDPR data protection impact assessment. Certain deployers, such as
public bodies, must also carry out a fundamental rights impact assessment (Art. 27).
General-purpose AI models
- All GPAI providers (Art. 53): technical documentation, information for downstream providers,
a copyright compliance policy and a public summary of training content. Free and open-source
models keep only the copyright and training-summary duties, unless they carry systemic risk. - Systemic risk (Art. 55): presumed above 10^25 FLOPs of cumulative training compute; the
provider must notify the Commission within two weeks. Extra duties: model evaluation including
adversarial testing, risk assessment and mitigation, serious incident reporting, cybersecurity. - The GPAI Code of Practice is the voluntary route to show compliance.
Penalties (Art. 99)
| Infringement | Maximum fine, whichever is higher |
|---|---|
| Prohibited practices (Art. 5) | EUR 35 million or 7% of worldwide annual turnover |
| Most other obligations | EUR 15 million or 3% |
| Incorrect or misleading information to authorities | EUR 7.5 million or 1% |
For SMEs and start-ups the lower of the two figures applies. Since the Omnibus, the same lower-of
rule applies to small mid-caps for the second and third tiers.
Key dates
| Date | What applies |
|---|---|
| 2025-02-02 | Prohibitions, definitions, AI literacy (Art. 4) |
| 2025-08-02 | GPAI obligations; governance bodies |
| 2026-08-02 | Most rules; Art. 50 transparency; enforcement starts |
| 2026-12-02 | New prohibition on sexual deepfakes and child sexual abuse material |
| 2027-08-02 | One AI regulatory sandbox per Member State |
| 2027-12-02 | Annex III high-risk rules (moved by the Omnibus) |
| 2028-08-02 | High-risk AI in Annex I products (moved by the Omnibus) |
Relation to other frameworks
- GDPR applies in parallel whenever personal data is processed. See data-privacy-in-ai.
- ISO/IEC 42001 gives the management system that can hold the Act’s risk management, data
governance, documentation and quality management duties in one place. Certification against it
does not by itself prove compliance with the Act. - NIST AI RMF gives a risk vocabulary and suggested actions that map onto Arts. 9–15.
- Harmonised European standards, which give a presumption of conformity for high-risk systems, are
being written by CEN-CENELEC JTC 21.
Official tools
- AI Act Service Desk: the Commission’s
AI Act Explorer,
Compliance Checker,
FAQ (with a Digital Omnibus section),
high-risk classification guidelines
with summaries and examples,
and national resources. - Fair Rules for Artificial Intelligence:
a free one-hour course by Sitra on eOppiva, no login required (announced 2025-05-07).
Scope
The Omnibus changes to Art. 4 (AI literacy) and to registration are covered in
ai-regulation-and-policy. This note is not legal advice.
Sources (accessed 2026-11-10)
- AI Act Service Desk, implementation timeline: https://ai-act-service-desk.ec.europa.eu/en/ai-act/timeline/timeline-implementation-eu-ai-act
- AI Act high-level summary (Future of Life Institute): https://artificialintelligenceact.eu/high-level-summary/
- Article 26: https://artificialintelligenceact.eu/article/26/
- Article 99: https://artificialintelligenceact.eu/article/99/
- Sitra course announcement: https://www.sitra.fi/en/news/a-new-online-course-supports-ai-literacy-and-clarifies-eu-regulations/