Canadian AI Register Omits Sociotechnical Context in 86% of Systems
Analysis finds Canada's AI Register frames systems as reliable internal tools while omitting human oversight and risk contexts, mirroring transparency shortfalls in EU and US frameworks.
Canada released its first Federal AI Register in November 2025 documenting 409 systems.
The arXiv paper by Dipto Das et al. applied the ADMAPS framework to the full dataset and found 86% of entries describe internal efficiency deployments while omitting human discretion, model training details, and uncertainty management practices (arXiv:2604.15514). The register's entries privilege technical specifications over sociotechnical descriptions of contestable decision-making.
Original coverage of the register launch did not address the systematic exclusion of high-risk system impact assessments required under Canada's Directive on Automated Decision-Making. The EU AI Act's public database similarly depends on self-reported summaries that independent audits have shown under-document prohibited and high-risk uses (European Commission AI Act Implementation Report, 2024). NIST's AI Risk Management Framework adoption data reveal comparable voluntary reporting gaps in the United States where high-impact federal uses receive limited external review (NIST AI 100-1, 2023).
AXIOM: Canada's registry lists hundreds of systems but leaves out how humans actually oversee high-risk decisions; the same design choice is appearing in EU and US guidance and will likely limit meaningful oversight worldwide.
Sources (3)
- [1]Bureaucratic Silences: What the Canadian AI Register Reveals, Omits, and Obscures(https://arxiv.org/abs/2604.15514)
- [2]EU AI Act: First Implementation Report(https://artificialintelligenceact.eu/implementation-report-2024)
- [3]Artificial Intelligence Risk Management Framework (AI RMF 1.0)(https://www.nist.gov/itl/ai-risk-management-framework)