The EU issues guidelines to help AI models with systemic risks comply with the AI Act, after criticism from some companies about lack of clarity
The European Commission set out guidelines on Friday to help AI models it has determined have systemic risks and face tougher obligations …
Context & Ripple Effects
The AI Act’s treatment of foundation models established a separate, stricter track for models designated as posing systemic risk, building on the framework described in the Act’s foundation-model restrictions. The Commission has since been translating the law into operating guidance, including guidance on prohibited AI practices.
This guidance follows the EU’s recent voluntary compliance code for AI Act obligations and addresses company complaints about how the systemic-risk rules should be applied. It matters because the focus is shifting from defining the rules to making them executable for the most consequential model providers.
First-order effects
- Providers of AI models deemed to carry systemic risk receive a clearer compliance reference point for the AI Act’s tougher obligations.
- The Commission reduces some immediate uncertainty for affected companies, while making their compliance expectations more concrete.
Second-order effects
- Model developers and their governance, safety, and documentation teams will need to align internal controls and evidence gathering to the Commission’s interpretation, rather than relying on broad statutory language.
- Clearer expectations can narrow the advantage of delaying compliance: competitors developing comparable models face a more legible path for building operational assurance.
Third-order effects
- If guidance continues to fill in the AI Act’s broad categories, EU AI regulation will increasingly function through detailed implementation materials alongside the statute itself.
- The episode reinforces a two-tier governance model in which the largest or most consequential general-purpose models face more formalized risk-management expectations than the wider AI market.
The trend: AI governance is moving from principle-level legislation toward operational compliance systems tailored to the risks posed by frontier-scale models.