Summary

• The government finder contained 142 Algorithmic Transparency Recording Standard entries when checked on 17 August.
• Records cover systems from local-authority contact-centre AI to education, recruitment, policing, health and central government tools.
• Central discoverability improves scrutiny, but the register still depends on public bodies identifying systems, updating records and supplying enough detail to assess them.

The Government Digital Service, Cabinet Office and Department for Science, Innovation and Technology have brought public-sector algorithm records into a single searchable service, giving automated systems used across government a more visible common register.

The Algorithmic Transparency Records finder contained 142 entries when checked on 17 August, spanning central departments, councils, policing bodies, NHS England and other public organisations. Records can be filtered by organisation, function, technical capability, deployment phase and region.

Recent entries show how varied the underlying systems have become. Newcastle City Council has disclosed production use of Amazon Q and Contact Lens in contact-centre services, while the Standards and Testing Agency has registered a large-language-model tool for generating KS2 writing examples and the Department for Education has recorded an AI service intended to check apprenticeship vacancies before publication.

The records range from production systems to pilots and tools still awaiting deployment, which makes the finder more useful than a catalogue of AI products already affecting services. It can show where organisations are experimenting as well as where algorithmic functions have become part of ordinary operations.

A register creates a trail, not an audit

Publishing an entry does not demonstrate that the technology is accurate, fair, secure or effective. The standard creates a structured account of what an organisation says a system does, why it is being used and how human oversight is arranged, giving scrutiny a firmer starting point without replacing independent assurance.

That distinction becomes more important as generative AI enters public services. A conventional rules-based system can often be described in fairly stable terms, whereas a large language model can produce variable outputs and may change when the underlying vendor updates it. Departments therefore have to keep records current rather than treating publication as a one-off compliance task.

Procurement can complicate disclosure as well. Public bodies using proprietary platforms need enough access to explain material aspects of an automated process, while suppliers may regard model details, security arrangements or product behaviour as commercially sensitive. Contracts consequently influence how much information an organisation can publish after a tool has been bought.

Those issues are particularly visible where AI is embedded into broader enterprise software. An organisation may purchase a contact centre, case-management platform or productivity suite and later activate algorithmic features inside it, making the boundary around what deserves a transparency record less obvious than when an AI system is procured separately.

Coverage will decide how useful the service becomes

The finder already spans a broad group of public organisations, although a searchable service can only expose systems that enter the reporting process. Central government has stronger disclosure requirements than parts of the wider public sector, leaving the possibility that similar technologies are described consistently in one organisation and less visibly in another.

Local government, policing, health and contracted service delivery make that completeness question significant because many high-consequence interactions with the state occur outside Whitehall departments. The register becomes more valuable as use of the standard spreads across those settings and as retired or materially changed systems are updated rather than left appearing current indefinitely.

Quality matters alongside quantity. A detailed record that explains data sources, human review and known limitations provides more useful accountability than a generic description of an “AI-enabled” service, even if both technically satisfy a disclosure process.

Centralising the records removes one practical obstacle by making them easier to discover and compare. The harder work now sits with the organisations creating them: keeping entries complete enough to withstand scrutiny while public-sector adoption of algorithmic systems continues moving from bounded pilots into services used every day.