EU AI Act (2025) - Key Takeaways
The key takeaways from the EU AI Act (2025) - the world’s first comprehensive regulatory framework for artificial intelligence:
Purpose and Scope
- Goal: To ensure AI in the EU is safe, transparent, and respects fundamental rights, while fostering innovation.
- Applies to: All AI systems that impact people in the EU, regardless of where they are developed or deployed
Risk-based Classification
AI systems are categorised into four risk levels:
- Unacceptable Risk – Banned outright (e.g., social scoring, real-time facial recognition in public)
- High Risk – Subject to strict obligations (e.g., in healthcare, finance, education, law enforcement)
- Limited Risk – Must provide transparency (e.g., chatbots must disclose they are AI)
- Minimal Risk – No restrictions (e.g., spam filters, video game AI)
Key Requirements
- Transparency & Accountability: High-risk AI must be explainable, auditable, and free from bias.
- Human Oversight: Critical decisions must involve human judgment.
- Data Governance: High-quality, representative data must be used to train AI systems.
- Documentation & Monitoring: Developers must maintain logs, undergo audits, and meet EU reporting standards
Compliance Timeline
- In force: From 1 August 2024.
- Phased rollout: Starting with bans on prohibited AI in February 2025, with full implementation by 2030
Penalties
- Non-compliance can result in fines up to €35 million or 7% of global annual revenue, whichever is higher
Strategic Goals
- Reduce EU dependence on non-European AI technologies.
- Set a global benchmark for responsible AI, similar to how GDPR shaped global data privacy norms