The Netherlands Court of Audit audited 9 government algorithms in 2022, finding only 3 met basic requirements, then made algorithm audits a recurring part of its annual audits through 2024 — a practice other audit institutions have since referenced.
3 of 9 algorithms audited
Algorithms meeting all basic requirements tested (2022)
9 algorithms
Algorithms audited in the initial review (2022)
3 algorithms
Further risk-prediction algorithms audited (2024 accountability audit)
Details
Maturity
Established
Promoter
Netherlands Court of Audit (Algemene Rekenkamer)
Period
2021–2024
Region (NUTS)
NL36
Keywords
public sector oversight, algorithmic accountability, government audit
Context
The Netherlands Court of Audit (Algemene Rekenkamer), the country's independent supreme audit institution, published 'An Audit of Algorithms' on 18 May 2022, assessing nine algorithms used across different Dutch government bodies against a structured framework covering governance, model quality, data quality and IT-general controls.
Objectives
Assess whether government algorithms meet basic requirements for governance, model quality, data quality and IT-general controls, and institutionalise algorithm auditing as a recurring element of the Court's oversight of central government.
Activities
The Court audited nine algorithms in its 2022 report; covered algorithms at the Ministries of Justice and Security and of Social Affairs and Employment in its 2022 results published in 2023; and audited three further risk-prediction algorithms across different government bodies in its 2024 accountability audit.
Results
Only three of the nine algorithms met all the basic requirements tested; the other six showed gaps ranging from inadequate performance monitoring to bias risks, data-leak exposure and insufficient control over unauthorised access. The approach and findings have been referenced internationally, including in a 2025 newsletter update from the EUROSAI IT Working Group, and the Court's audit framework has been used by outside researchers to re-examine specific government risk-profiling systems.
Conclusions
This is a governance/oversight practice rather than a deployed citizen-facing AI tool. It is included as strong evidence of a replicable, sustained institutional mechanism for holding public-sector AI accountable, even though the 2022 findings were largely critical of the audited algorithms.
Implementation
Indicative cost
Low (< €50k)
Time to results
Long (> 3 years) — Recurring annual audits: 2022 (9 algorithms), 2023 results covering the Ministries of Justice and Security and of Social Affairs and Employment, 2024 (3 further risk-prediction algorithms).
Staffing & skills
Netherlands Court of Audit (Algemene Rekenkamer) audit teams applying a structured framework covering governance, model quality, data quality, and IT-general controls
Conditions for success
Institutionalising algorithm audits as a recurring element of the annual accountability audit rather than a one-off review
Publishing full audit findings and methodology for public and international scrutiny
Common failure modes
Six of nine audited algorithms showed gaps ranging from inadequate performance monitoring to bias risks, data-leak exposure and insufficient control over unauthorised access
Commonly funded by
National / regional programmes
Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.
Replication kit
Reusable artefacts from this practice — as published by their sources.
Banco de España's Spain-STING model nowcasts quarterly Spanish GDP growth in real time from early indicators; a 2024 respecification (contemporaneous …