Summary
The European Union AI Act enters full enforcement with new requirements for AI systems including mandatory transparency, risk assessments, and documentation standards.
Key Takeaways
- 1Full enforcement began May 2026 with penalties in August
- 2High-risk systems require documentation, logging, human oversight
- 3Penalties up to 35M EUR or 7% global revenue
- 4Applies to any company affecting EU citizens
- 5Transparency required for chatbots and generated content
The EU AI Act is now fully enforced, and if your company uses AI, you need to understand these requirements.
**Who Is Affected**
Any company deploying AI systems in the EU market, companies outside EU if their AI affects EU citizens, AI providers, deployers, importers, and distributors.
**Risk Categories**
Prohibited AI includes social scoring systems and real-time biometric identification (with exceptions). High-risk AI includes recruitment tools, credit scoring, educational assessment, medical diagnosis, and law enforcement applications. Limited risk requires transparency for chatbots and deepfakes. Minimal risk includes spam filters and video games.
**Documentation Requirements**
For high-risk systems: technical documentation covering design, data, and testing; risk assessment and mitigation measures; logging capabilities for traceability; instructions for use; human oversight measures; and performance monitoring procedures.
**Compliance Timeline**
May 2026: All provisions now in force. August 2026: Penalties begin (up to 35M EUR or 7% of global revenue). Ongoing: Continuous monitoring and documentation required.
**Practical Steps**
Inventory all AI systems, classify by risk category, document data sources and training procedures, implement logging for AI decisions, establish human oversight procedures, create incident response plans, and train staff on compliance requirements.
Beginner Friendly
The EU AI Act is like a safety inspection for AI systems. Just as cars must meet safety standards, AI systems must now meet transparency and safety requirements before they can be used in Europe.
Advanced Insights
From a legal engineering perspective, implement immutable audit logs for AI decisions, model cards for documentation, human-in-the-loop approval workflows, and automated compliance checking in your CI/CD pipeline.
Frequently Asked Questions
Quick answers about this story
Thomas Wright
AI Writer & Researcher
Reviewed by OneStep AI editorial team