Algorithm register vs AI register vs the EU database

Three registers, three jobs

• Public algorithm registers: transparency websites where governments publish the algorithms they use to make decisions about citizens. Amsterdam and Helsinki pioneered municipal registers in 2020; the Netherlands runs a national Algoritmeregister where government bodies publish impactful algorithms; the UK's Algorithmic Transparency Recording Standard plays a similar role for its public sector.
• The EU AI Act database: the EU-level registration system under Article 49. Providers register high-risk Annex III systems before placing them on the market, and deployers that are public bodies register their use of such systems. Most of it is publicly viewable.
• Your internal AI register: the operating record of every AI system you provide or deploy — classifications, roles, owners, evidence. Nobody sees it unless you show them; nothing in your program works without it.

The three share a word and almost nothing else: different audiences (the public, the regulator, your own organization), different scopes (government algorithms including non-AI rule-based systems, high-risk Annex III systems only, your entire AI portfolio), and different legal footings (national transparency policy, a binding EU-level duty, and operational necessity).

Which ones actually apply to you

For a private product company, the answer is usually: the internal register always, the EU database sometimes, the public algorithm registers almost never — but each deserves the check.

• Internal AI register: applies to everyone. Every per-system duty in the AI Act — Article 5 screening, Article 6 classification, Article 26 deployer obligations, Article 50 transparency — presupposes it.
• EU database (Article 49): applies when you are the provider of a high-risk Annex III system — registration happens before market placement, alongside conformity assessment. Providers who concluded a system is not high-risk via the Article 6(3) derogation also register that assessment. As a deployer, the duty to register use applies to public authorities and bodies acting on their behalf — a private SaaS deployer does not file its use there.
• Public algorithm registers: apply to government bodies in the jurisdictions that run them. Relevant to you if you sell to those bodies — your public-sector customers may need inputs from you to publish their entry, and being easy to publish about is a quiet procurement advantage.

Why the Dutch keep asking about this

If you searched "AI register opzetten" — setting up an AI register — you are in the jurisdiction where this distinction is most alive. The Netherlands runs the most developed national algorithm register, Dutch supervisory authorities have been vocal about algorithmic transparency, and Dutch organizations consequently treat an internal AI register as standard governance hygiene rather than an exotic compliance artifact. The practical setup advice is the same everywhere: inventory your systems, classify each one, name owners, attach evidence, and review on a rhythm — the register is the record that rhythm produces.

One record, three outputs

The efficient architecture is not three parallel documents. It is one maintained internal register, from which the other two are projections: when a system becomes a high-risk provider situation, the Article 49 filing is an export of fields you already keep; when a public-sector customer needs material for their algorithm register entry, that is an export too. Organizations that maintain the internal record produce the public-facing ones in an afternoon; organizations that do not, reconstruct history under deadline.

Frequently asked questions

Keep reading

• What is an AI register — and why every deployer needs one
• How to build an AI system inventory, step by step
• What actually applies on August 2, 2026