Algorithmic Documentation and the Limits of Freedom of Information: The EU AI Act's Two-Track Disclosure Architecture
Keywords: freedom of informa7on, AI Act, automated decision-making, GDPR, algorithmic transparency, government accountability, FOIA exemptions, public records disclosure
Abstract: Freedom of information law and EU AI governance law share a common premise: government accountability requires access to
government information. In practice, they pull in opposite directions. FOIA regimes are reactive: they give citizens the right to request documents after the fact. The EU AI Act is prospective: it requires technical documentation before a high-risk system is deployed. When a FOIA request targets an AI system's decision logic, these two regimes collide. The resulting disclosure asymmetry is documented across five welfare AI systems in EU member states, each a case where public access mechanisms failed to produce algorithmic documentation or performance information that the AI Act would require to exist within the supervisory compliance architecture. The AI Act creates a two-track disclosure architecture: supervisory access to technical documentation is legally enforceable, while public access to model-level documentation is not. For FOIA purposes, algorithmic documentation remains vulnerable to withholding
under fraud-prevention, public-control, and commercial-confidentiality exemption grounds in the access-to-documents disputes examined, while adjacent regulatory-disclosure mechanisms likewise failed to produce public model-level documentation. This architecture privileges regulatory insiders over citizens. The paper derives a minimum disclosure standard from the AI Act's existing obligations, one that closes the gap without requiring legislative amendment in its core elements.
Email Sharing: We authorize the sharing of all author emails with Program Chairs.
Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.
Submission Number: 5
Loading