Denmark - Algorithm Transparency Proposal (B 136)

Overview

Proposal B 136 (fremsat 29 February 2024) requests that the Danish Government table legislation to ensure that public authorities' use of algorithms and automated decision‑making in administrative casework is governed by principles of transparency and explainability. The decision proposal, authored by members of the Socialistisk Folkeparti, notes recent cases where citizens could not access the models, variables or calculations underpinning public decisions and asks for statutory safeguards so parties can know whether algorithms were used and what data informed outcomes. The parliamentary dossier and supporting materials are available on the Folketinget website (B 136 overview) and the submitted text (B 136 som fremsat). The proposal ultimately failed at the second reading but remains a key reference for transparency and public sector AI governance debates in Denmark. (ft.dk)

Definitions

For the purposes of the proposal and this summary, key definitions are: "algorithm" — any computational procedure, model or code (including machine learning models) that processes data and contributes to administrative decisions; "automated decision‑making" — decisions or assessments where algorithmic outputs materially inform or determine outcomes for a citizen; "transparency" — provision of meaningful information to affected parties about the use, purpose, and functioning of algorithmic tools; "explainability" — the ability to provide an intelligible account of how algorithmic outputs relate to decisions in a specific case; "oversight bodies" — public auditors and accountability institutions (e.g., Rigsrevisionen, Folketingets Ombudsmand, Datatilsynet) with statutory powers to inspect systems. These working definitions align with the framing used in the submitted proposal text. (ft.dk)

Governance and Institutional Framework

The proposal assigns primary responsibility for drafting enabling legislation to the Government (Ministeriet for Digitalisering og Ligestilling) and envisages a governance architecture in which existing oversight institutions — notably Folketingets Ombudsmand and Rigsrevisionen — are explicitly empowered to access algorithmic models and related documentation in exceptional circumstances to carry out independent audits. Operational responsibilities for compliance would rest with the individual authority using the system: each authority would have to document systems, perform impact assessments, and maintain logs. The proposal also anticipates interplay with the Danish Data Protection Agency (Datatilsynet) on personal data issues and with central procurement and IT governance frameworks for public sector systems. The Folketing documentation lists the Ministerial area responsible as the Ministry for Digitalisation and Equality and details the committee referral and report stages. (ft.dk)

Key Focus Areas

B 136 concentrates on several policy priorities: (1) Procedural transparency — ensuring that when algorithmic methods are used in individual decisions, case parties are informed and able to obtain concrete information about which data and variables were determinative; (2) Explainability and right to an intelligible justification — authorities must provide meaningful explanations permitting the individual to understand the basis of the decision; (3) Oversight access — in limited cases where commercial secrecy or other legitimate confidentiality concerns would otherwise block disclosure, parliamentary auditors and the Ombudsman's office should be granted access to ensure proper control; (4) Data governance — the proposal stresses compliance with data protection rules (including GDPR and Danish data protection law) and suggests that authorities should prefer open source or maximally transparent implementations for public sector systems; (5) Impact assessment and documentation — authorities deploying algorithmic systems should perform and publish assessments (including rights and fairness impacts) before operational use; (6) Human oversight and appealability — there should remain a clear route for human review and for redress when algorithmic processes affect fundamental rights; and (7) Procurement and technical architecture — public procurement and system design should favour solutions that enable scrutiny, reproducibility and independent auditability. These focus areas reflect both the text of the proposal and the parliamentary debate that accompanied the bill's readings. (ft.dk)

Implementation Framework

The proposal envisions implementation by a combination of legislative amendment, administrative rules, and guidance. Core measures would include amendments to the Danish Administration of Justice Act (forvaltningsloven) to incorporate explicit transparency and explainability obligations, statutory requirements for pre‑deployment impact assessments for algorithmic systems in the public sector, mandatory documentation and logging standards, and protocols for secure disclosure to oversight bodies. The draft envisaged specifying procedures for balancing disclosure against legitimate confidentiality (e.g., national security, trade secrets, or privacy of third parties) and for enabling oversight bodies to examine models, training data and code under confidentiality agreements where necessary. Operationally, authorities would need to integrate documentation into case management systems, maintain reproducible artefacts for audit, and designate responsible officers for algorithmic systems. The proposal also contemplates central guidance and possibly a registry of publicly used algorithmic systems to support transparency. (ft.dk)

Monitoring and Evaluation

Monitoring would combine internal compliance reviews by authorities, independent audits by Rigsrevisionen and inspections by Folketingets Ombudsmand and Datatilsynet, and public reporting obligations. The proposal suggests that oversight bodies should be able to require documentation and test data and, where needed, deploy expert technical review. Evaluation criteria would include correctness, fairness (non‑discrimination), explainability, data provenance, repeatability of outputs, and procedural conformity with administrative law. Regular public transparency reports and a central register would enable external researchers and civil society to monitor deployment patterns and systemic risks. The parliamentary record recommends that monitoring focus particularly on systems with large scale or high impact on fundamental rights. (ft.dk)

Penalties, Liability, and Appeals

As a decision proposal requesting that the Government draft a bill, B 136 did not itself set binding enforcement sanctions; however, it anticipated that implementing legislation should include remedial and enforcement mechanisms. Possible mechanisms discussed in the parliamentary debate include administrative orders to suspend or modify systems, mandatory remediation plans, public disclosure of non‑compliance, financial or reputational penalties, and strengthened powers for oversight bodies to compel disclosure. The proposal also emphasises the importance of preserving individual appeal rights under administrative law and ensuring that citizens can access effective remedies if algorithmic systems produce incorrect or unlawful outcomes. Any enforcement regime would need to be balanced against proportionality, legal certainty and international obligations like EU law and GDPR. (ft.dk)

Relationship to Other Instruments

B 136 situates itself alongside existing Danish and EU instruments. Domestically it builds on principles in the Danish Administration of Justice Act (forvaltningsloven) and the Danish Data Protection Act (databeskyttelsesloven) as implemented under the GDPR. It also complements prior parliamentary proposals and debates (e.g., proposals on AI oversight and data ethics) referenced in the dossier. Internationally, the proposal anticipates alignment with EU‑level developments (notably the EU AI Act and GDPR obligations regarding automated decision‑making and transparency). The parliamentary materials cross‑reference earlier proposals from the same term addressing independent AI oversight and risk assessments. The text therefore should be read as a national-level proposal to operationalize judicial and administrative transparency norms in the context of rapidly advancing algorithmic public systems. (ft.dk)

International Alignment

Although B 136 is a domestic decision proposal, its norms are crafted with EU developments in mind. The proposal's emphasis on risk‑based assessment, explainability, and oversight is consistent with the EU AI Act's approach to classifying high‑risk systems and requiring conformity and transparency measures. It also reflects GDPR concerns about automated decision‑making affecting individuals and data protection impact assessments (DPIAs). The proposal therefore aims for a Danish regulatory response that would be coherent with emerging EU obligations and existing international standards on algorithmic accountability, while preserving the roles of national oversight bodies. Policymakers and drafters would need to consider the interplay between national transparency measures and EU restrictions on trade secrets and IP when designing mandatory disclosure or audit access. (ft.dk)

Implementation Timeline

Sources and References