EU AI Office: Post-Release Monitoring for Advanced AI Models

The EU AI Office has outlined how advanced AI models will be monitored after their release. This process aims to strengthen assessments of systemic risks that might not be fully identified before deployment. This issue was discussed at a recent meeting of the taskforce for the EU’s voluntary General-Purpose AI Code of Practice. The discussion covered the code’s Safety and Security and Copyright chapters. The AI Office believes post-deployment information on model use and behavior can inform systemic risk assessment. Participants also looked at "marginal-risk" clauses, suggesting they should only be used in exceptional cases. On copyright, the meeting addressed commitments for providers to publish web crawler information and support robots.txt to respect rights reservations. The code remains a voluntary way for providers to show compliance with the EU AI Act. This matters because monitoring deployed models could reveal risks missed earlier, and greater transparency helps rightsholders understand how AI providers respect their content.