A new analysis reported by arXiv cs.AI has examined the state of government transparency regarding artificial intelligence. The study compared 8,368 individual records from AI inventories and registers across 72 countries, as well as transnational databases. These inventories are designed to make the use of AI within public sectors visible to the public, yet the research suggests that current reporting practices are inconsistent and often incomplete.
The Challenge of Inconsistent Reporting
The researchers evaluated the data across 23 harmonized fields to determine how information is structured. While most registers shared a basic descriptive core—identifying that an AI system exists and its general purpose—they frequently omitted critical details. Information regarding the legal basis for using a specific tool, the results of external evaluations, and the potential risks posed by the technology was rarely requested or provided. Furthermore, the study noted that when governments used broad reporting schemas, they often resulted in significant amounts of missing data. This lack of detail hinders the ability of citizens and oversight bodies to understand how AI influences public policy and administrative decisions.
Unknowns in Global AI Governance
The reporting did not specify which specific countries or jurisdictions maintained the most comprehensive registers, nor did it rank the analyzed inventories by quality. While the study highlighted a lack of patterned convergence in how these schemas are designed, it did not provide a recommended universal standard for governments to adopt. It remains unclear how individual agencies intend to address the missing data in their current records or if there are plans to update these registers to include information on appeals processes for citizens affected by automated decisions. The announcement also did not specify the exact timeframe during which these 8,368 records were collected or if any specific AI models were more prevalent in certain regions.
The findings suggest that while the infrastructure for AI transparency exists in many nations, the actual content of these registers lacks the interoperability needed for global comparison. Without more rigorous reporting on risks and legal frameworks, the effectiveness of these inventories as tools for public accountability remains limited.
Lead image: Christina Morillo / Pexels
