EU Regulatory Center
EU AI Act — risk-based rules for AI systems
A risk-tiered framework governing AI systems placed on the EU market, with obligations that phase in by risk category and role.
Status
Law in force
Regulation (EU) 2024/1689
The AI Act entered into force in August 2024 and applies obligations in phases by risk tier: prohibited practices, transparency duties for general-purpose AI, and the fuller set of obligations for high-risk AI systems apply on a staggered timeline extending into 2026 and beyond.
Scope at a glance
| Question | Answer |
|---|---|
| Who | Providers, deployers, importers and distributors of AI systems placed on or used in the EU market. |
| What | Risk-tiered obligations: unacceptable-risk practices are prohibited; high-risk systems face conformity and governance duties; limited-risk systems face transparency duties. |
| Governed AI | Systems that interact with people — such as CYRKIL’s own LISA — fall under transparency obligations requiring clear AI disclosure. |
| Timeline | Obligations are phasing in on a staggered schedule by risk tier and role, rather than a single application date. |
Core obligations
- Classify AI systems by risk tier and apply the corresponding obligations.
- Disclose clearly when a person is interacting with an AI system rather than a human.
- Maintain governance, documentation and human-oversight measures for high-risk systems.
- Monitor and report serious incidents involving high-risk AI systems where required.
General guidance, not legal advice
This page summarizes publicly available regulatory status for general guidance only. Confirm applicability, scope and deadlines with qualified legal counsel before making compliance decisions.
How CYRKIL applies this to itself
LISA, CYRKIL’s own AI guide, is built around the AI Act’s disclosure principle — always identified as an AI system, never presented as human.
See LISA’s disclosure modelHow CYRKIL tracks this
Every regulatory page here runs through the Regulatory Truth Pipeline — official source, human and legal review, then publish, with the source, version and review date always shown.
See the pipeline