Researchers expose gaps in U.S. federal AI transparency

A new paper titled *Triangulating Across U.S. Federal AI Transparency Regimes* (arXiv:2607.29540) reveals systemic failures in how the U.S. government discloses its AI systems. Researchers Emma Lurie, Emma Fauser, Qing He, Danaé Metaxa, and Sorelle Friedler examined three transparency mechanisms: System of Records Notices (SORNs), Information Collection Requests (ICRs), and the AI Use Case Inventory. Their analysis found that no single regime provides complete visibility into federal AI deployments, and the lack of persistent identifiers makes it nearly impossible to track systems across documents or over time.

The team developed a triangulation method using zero-shot classification and cross-document entity resolution to link disclosures across regimes. While this approach improved insight into AI use cases, they conclude that even linked records fall short of meaningful oversight. For example, the AI Use Case Inventory’s annual reporting cycle allows agencies to deploy AI systems months before they appear in official records. The researchers attribute these gaps to the regimes’ original administrative purposes and propose reforms, including standardized AI definitions, persistent identifiers, and restored public access to risk management processes.