Explainability and transparency of government algorithms: FRIA and the register

Short answer: Transparency of government algorithms has two axes. Collective: openness about which systems run, via the algorithm register and the fundamental rights assessment (FRIA, Art. 27 AI Act). Individual: explanation to the specific citizen about a decision that affects them, via the administrative-law duty to give reasons and the GDPR. The AI Act additionally requires high-risk systems to be intelligible and logged. Explanation is a legal duty, not a favour.

Two kinds of transparency

It is crucial to separate collective from individual transparency. Collective transparency answers which algorithms a government uses and for what purpose — that is public accountability. Individual transparency answers why my decision came out as it did — that is legal protection. An entry in the register does not replace individual reasoning, and vice versa. Both are needed.

The algorithm register

The algorithm register is the collective transparency instrument: a public overview of algorithms governments deploy, with information on purpose, operation, data and safeguards. It enables external supervision and societal scrutiny. For high-risk systems this connects to the EU database under the AI Act. Registration is not an endpoint but the starting point of reviewability.

The FRIA as a structured assessment

The fundamental rights assessment — see FRIA: Art. 27 — forces a government to map, before use, which fundamental rights are at stake, which groups are affected and which safeguards exist. The FRIA is thus both an internal risk assessment and a transparency product: its outcome feeds the register and substantiates deployment. See also AI in the public sector and algorithmic decision-making.

Explainability is both technical and legal

The AI Act requires high-risk systems to be designed so their operation is sufficiently transparent (Art. 13) and that events are logged (Art. 12). Technically this means reconstructable outcomes and traceable factors. Legally it means a decision must be properly reasoned (administrative law) and the citizen has a right to meaningful information (GDPR). A model that admits no explanation is simply unusable for government decisions with legal effect.

What to do

• Separate collective and individual transparency: deliver both, not one.
• Register in time in the algorithm register and the EU database for high-risk systems.
• Run the FRIA before use and use its outcome as a transparency product.
• Secure logging and reconstructability (AI Act Arts. 12-13), so decisions can be reasoned and hold up.
• Give the citizen meaningful explanation of decisions that affect them, with the ability to contest.

For government algorithms, transparency is not a communications exercise but the precondition for lawful deployment. What you cannot explain, you may not decide.