EU AI Act: All You Need to Know
The EU Artificial Intelligence Act, adopted in June 2024, is the world’s first comprehensive legal framework for AI. Designed to ensure that artificial intelligence in Europe is trustworthy, safe, and human-centric, the Act also includes special provisions to support SMEs and startups navigating the complex AI landscape. It introduces a risk-based system with three tiers:
Unacceptable risk systems (like real-time biometric surveillance or social scoring) are banned
High-risk systems (used in hiring, education, healthcare, or law enforcement) require conformity checks and registration
Limited-risk and general-purpose AI tools face lighter obligations—mostly transparency rules
Startups stand to benefit from ready-made Codes of Practice, and early access to regulatory sandboxes, rare provisions that level the playing. SMEs are protected from excessive fines under the Act’s proportionality principle, which ensures penalties are scaled to company size and financial capacity, reducing compliance risk for early-stage ventures.
One impressive demonstration of this principled, public-interest approach is Apertus, Switzerland’s fully open-source AI model, launched in September 2025 by EPFL, ETH Zurich, and the National Supercomputing Centre. Available in both 8-billion and 70-billion parameter versions via Hugging Face and Swiss partners, its architecture, weights, training data, and documentation are openly accessible.
Designed to meet both Swiss and EU data and copyright standards, Apertus supports over 1,000 languages, including Swiss German and Romansh, marking a breakthrough in transparent, compliant, foundational AI. It stands as a timely example of how regulation can foster responsible innovation. For European AI-driven SMEs, the message is clear: the regulatory ceiling is rising, but so are public, trustworthy models and compliance tools.
What You Should Do in 2025
🔸 Stop any projects that may fall under the banned “unacceptable risk” categories
🔸 Train your team in basic AI literacy, understanding what the Act regulates and why
🔸 Read and follow the upcoming Codes of Practice as a shortcut to compliance
🔸 If you build with generative AI, prepare disclosure statements and copyright notices
🔸 Engage with your national AI authority to explore sandboxes, pilots, or get questions answered early.
Looking Ahead
Full implementation of the AI Act is expected by mid-2026, but 2025 is the critical year to act. With governments offering clear rules, support tools, and dedicated contact points for smaller players, there’s no better time to align, adapt, and build.