Public algorithm register

A public algorithm register is an official, publicly accessible list of the algorithmic and AI systems that public bodies use to make or support decisions. For each system it typically records what the system does, why it is used, which data it relies on, who is responsible, what human oversight exists, what risks were identified and how people can complain. The aim is to make the use of automated systems in government visible before problems occur, rather than discovered through scandals, litigation or investigative journalism.

Registers are a transparency mechanism, not a control mechanism. They do not by themselves approve or prohibit a system, but they create the documentation that ombudsmen, auditors, journalists, researchers and affected people need in order to scrutinise it.

Three lines of argument support registers.

1. Democratic accountability. Citizens cannot contest what they do not know exists. Governments are increasingly using algorithms to allocate benefits, target inspections, prioritise cases and detect fraud, decisions that once required human judgement and published rules. A register restores something like the publicity of law for automated rules.
2. Algorithmic legitimacy. Trust in automated public decisions depends on perceived fairness, explicability and the possibility of appeal. Publishing the existence and purpose of systems is a precondition for each.
3. Administrative learning. A register forces public bodies to inventory their own systems, which many have never done, and allows comparison and reuse across agencies.

Critics of "transparency theatre" warn that disclosure without explanation, enforcement or capacity can legitimise systems while changing little. Floridi, discussing the first city registers, framed them as a step towards treating AI as a public service subject to public standards.

What a register entry usually contains

• Amsterdam and Helsinki (2020). On 28 September 2020 the two cities announced open AI registers, described as the first offered by any city, listing ordinary systems dealing with issues such as parking and the reporting of maintenance problems. Amsterdam's own register has since been folded into the national Dutch register.
• The Netherlands. After the childcare benefits scandal, in which parents were wrongly branded as deliberate fraudsters and the tax administration's use of dual nationality in risk-selection models was found unlawful and discriminatory, the Dutch government launched a national Algorithm Register in December 2022. By October 2026 it listed 1,606 algorithms from central and local government bodies, with a focus on high-impact systems.
• United Kingdom. The Algorithmic Transparency Recording Standard (ATRS) provides a common template for public bodies to publish information on algorithmic tools. It is mandatory for all government departments and for arm's-length bodies that deliver public or frontline services or interact directly with the public, and recommended for the wider public sector.
• France. The 2016 Law for a Digital Republic requires administrations to explain the rules of algorithmic processing used in individual decisions on request and requires larger administrations to publish the rules of their main algorithmic treatments.
• European Union. The AI Act (2024) creates an EU database for high-risk AI systems; public authorities that deploy high-risk systems must register their use, extending the register logic across the Union.

• Incompleteness. Registers depend on self-reporting. Agencies may omit systems, define "algorithm" narrowly, or exempt law enforcement, migration and security uses, which are often the most consequential.
• Superficial entries. Entries can be vague or written in promotional language, offering little that an affected person or researcher can use.
• No link to rights. Without a right to explanation, appeal or human review, knowing that a system exists does not help individuals contest outcomes.
• Procurement opacity. Systems bought from vendors may be protected by trade secrets, limiting what can be disclosed.
• Maintenance burden. Registers must be updated as systems change; stale entries mislead.
• Legitimisation risk. Publication can make systems appear vetted when no assessment has occurred.

Algorithm registers spread from two city pilots in 2020 to national standards in the Netherlands and the United Kingdom and an EU-wide obligation for high-risk public uses under the AI Act. They have improved basic visibility of government automation and created a common vocabulary for describing systems. Evidence that they change outcomes, by preventing harmful systems or improving redress, is still limited, and their value depends on completeness, the quality of entries and the existence of complementary tools such as impact assessments, audits and appeal rights.

• Algorithm Register of the Dutch government
• UK Algorithmic Transparency Recording Standard Hub
• Floridi (2020), "Artificial Intelligence as a Public Service: Learning from Amsterdam and Helsinki", Philosophy & Technology
• Regulation (EU) 2024/1689 (AI Act)