The European Union's AI Act, which began phased implementation in 2025, has seen most of its provisions come into force by 2026. For the open-source AI community, the impact has been less severe than previously feared.
Open-Source Exemptions
The AI Act includes specific exemption clauses for open-source models—if you release a research-oriented open-source model, many compliance requirements do not apply.
However, there is a key exception: if your open-source model is used in "high-risk" application scenarios (such as medical diagnosis, credit assessment, law enforcement, etc.), the same rules must still be followed.
Practical Impact
For most open-source developers, daily operations have barely changed. Meta continues to release Llama weights, Hugging Face continues to operate its Hub platform, and small to medium-sized open-source projects do not need additional compliance work.
The biggest impact falls on the deployment side—if a company uses an open-source model in a high-risk scenario, the company bears the compliance responsibility. This effectively shifts the compliance burden from model developers to model users.
Comparison with China and the US
| Dimension | EU | US | China |
|---|---|---|---|
| Core logic | Risk-based classification | Industry self-regulation | State-led |
| Attitude toward open source | Exemptions + exceptions | Largely unregulated | Requires approval |
| Enforcement strength | Legally binding | Voluntary standards | Administrative orders |
Impact on Chinese Companies Going Global
Chinese AI companies looking to enter the European market must comply with the AI Act's requirements. This is not a major issue for large companies (which have legal teams to handle it), but it poses a barrier for small and medium-sized enterprises.
Conversely, China's own AI regulatory framework is in some respects stricter than the EU AI Act (for example, in content review requirements), so products that can operate smoothly under China's domestic compliance environment are not particularly difficult to adapt to EU requirements.
Overall: The AI Act is the world's first systematic AI regulatory law, and its demonstration effect may be more significant than its specific provisions. Other countries and regions will reference this framework when formulating their own AI regulations.
Sources: EU AI Act text, CocoLoop, Hugging Face policy analysis