The EU AI Act (Regulation (EU) 2024/1689) is the world's first comprehensive legal framework for artificial intelligence. It applies to any organisation that develops, deploys, imports, or distributes AI systems that affect people in the European Union: regardless of where the organisation is based.
If you're preparing for the AIGP exam, the EU AI Act is one of the most heavily tested topics. Here's what you need to know, in plain language.
The EU AI Act doesn't ban most AI: it classifies AI systems by risk level and applies proportionate obligations. The higher the potential harm, the stricter the rules. There are four tiers.
Banned outright. These systems pose an unacceptable threat to fundamental rights and human dignity.
Permitted but subject to strict obligations before they can be deployed. Must be registered in an EU database.
Minimal requirements: mainly transparency (e.g. telling users they're interacting with an AI).
Spam filters, AI in video games, recommendation engines. No regulation required under the Act.
These are banned from August 2026 onwards. They include:
Exam tip: A common trap is classifying social scoring as "high-risk" rather than "prohibited." It is banned outright: not regulated with extra requirements.
High-risk AI systems can be deployed but require substantial compliance work. High-risk categories include:
| Requirement | What It Means in Practice |
|---|---|
| Risk management system | Ongoing identification and mitigation of risks throughout the lifecycle |
| Data governance | Training data must be relevant, representative, and free from bias |
| Technical documentation | Full documentation before deployment and kept up to date |
| Transparency & instructions | Users must be given clear information about the system's purpose and limitations |
| Human oversight | Humans must be able to intervene and override AI decisions |
| Accuracy, robustness & cybersecurity | Systems must meet performance standards throughout their lifecycle |
| Conformity assessment | Third-party or self-assessment to verify compliance before market placement |
| Registration | Must be registered in the EU AI database before deployment |
General Purpose AI (GPAI) models: like large language models: have their own set of rules under the Act. All GPAI model providers must:
GPAI models deemed to present systemic risk (trained with over 10^25 FLOPs) face additional requirements including adversarial testing and serious incident reporting.
The Act has broad extraterritorial reach. It applies to:
This means a US company with no EU office is still subject to the Act if its AI affects EU citizens.
| Date | What Comes Into Force |
|---|---|
| August 2024 | Act enters into force |
| February 2025 | Prohibited practices provisions apply |
| August 2025 | GPAI model obligations apply |
| August 2026 | High-risk AI system obligations apply (most provisions) |
| August 2027 | Obligations for certain high-risk systems in Annex I apply |
The EU AI Act is one of the most tested areas in the AIGP exam. Focus on:
Practice with 200 scenario-based questions including a full EU AI Act domain: free, no payment needed.
Start Free Practice Quiz →