The Algorithmic Transparency Recording Standard (ATRS) is meant to give the British public a clear view of how the central government uses AI and algorithmic tools. It became mandatory for central government departments in February 2024. This page pulls every published record live from GOV.UK and shows what is actually there, organised so you can browse, filter and judge for yourself.
The headline question is whether the register is doing what it was set up to do. The figures below are an answer of sorts. So is the absence of an answer. The records that should be on the register, but are not, matter just as much as the ones that are.
The register at a glance
Growth has been recent and uneven
The ATRS launched as a voluntary pilot in November 2021 and produced a handful of records over the following two years. Compliance only became mandatory in February 2024, but for most of 2024 the register barely moved. The real growth came in 2025, driven by a Roadmap commitment to publish every in-scope tool by year end and by sustained pressure from the Public Accounts Committee.
Who is publishing, and what kinds of tools
Two breakdowns matter. Which organisations are using the register, and what categories of algorithmic tool they are reporting. A healthy register would show coverage across the full sweep of central government, and a realistic spread of tool types from low-risk calculators to high-impact decision systems. Neither pattern holds in the data below.
By organisation
By tool category
Categories are inferred from the title and description of each record using a simple keyword classifier. See methodology below.
What the register is not showing
The simplest test of a transparency register is whether the numerator matches the denominator. The ATRS gives us the numerator: the count of records published. The denominator is the actual scale of algorithmic tool deployment across the UK public sector. The two numbers do not line up.
The NAO figure is now over two years old and reflects only the bodies in scope of that survey. Total deployment across central government, arm's-length bodies, the NHS, police forces and local government is certainly several hundred tools and is growing quickly under the AI Opportunities Action Plan.
The register is not a list of everything the public sector is using. It is a list of what the public sector has chosen to declare. The mandatory scope policy published in December 2024 explicitly excludes large categories of use, including national security applications, broad policy analysis tools, and so-called "slipstream" AI that arrives bundled inside commodity software such as Microsoft Copilot.
Browse every record
Filter by organisation, category or year. Click any record to read the full transparency report on GOV.UK.
Three observations
The data above tells us things the register's own framing tends to soften. Here are three that matter for anyone trying to read the state of AI in British government.
The register favours the easy case
Scan the records and the pattern is unmistakable. Calculators, planners, chatbots and identity checks dominate. These are useful tools and worth publishing, but they are also the safest things to publish. The tools that genuinely concentrate power, where an algorithmic output shapes a decision about benefits, immigration, fraud detection or police priorities, are thinly represented. Transparency that operates only where the stakes are low is not the same as transparency.
Mandates without consequences do not move government
The trajectory is the proof. From February 2024, when the mandate became live, to late 2024 the register barely moved. Records arrived in volume only once departments faced direct accountability through the Public Accounts Committee and a public Roadmap commitment with a delivery date. The lesson generalises beyond the ATRS. Voluntary standards are easy to ignore. Mandatory standards without enforcement consequences are nearly as easy. The lever that actually worked was being named in a select committee report.
The denominator is unknown by design
Government can declare the register complete only because government decides what counts as in scope. The December 2024 scope and exemptions policy carves out national security, broad analytical tools, and tools that support decisions about "groups" rather than identifiable individuals. It also has no answer for the commodity AI now arriving inside every productivity tool the civil service buys. The result is a register that meets its own definition of completeness while telling us very little about the real shape of AI in British government. A genuine transparency regime would start with an independent inventory of tools, not with a self-reported list of what departments are willing to declare.
Methodology and caveats
Data source
Records are fetched live from the GOV.UK Search API, filtered by document type algorithmic_transparency_record. The page paginates through the full result set on load. The "last fetched" timestamp at the top reflects when your browser pulled the data.
Categories
Each record is assigned a tool category based on keyword matching against its title and description. The taxonomy covers chatbots, calculators and planners, risk scoring, identity verification, classification, recommendation, search and retrieval, document processing and forecasting. Classifications are approximate; some records sit across categories. Where the inference is genuinely ambiguous the record is tagged as "other".
Organisations
The organisation attribution is derived from the record's title prefix (e.g. "DSIT:", "DWP:", "HMRC:") rather than from the GOV.UK organisations metadata, because the metadata cross-publishes most records to the Cabinet Office, DSIT and GDS regardless of which department actually owns the tool. Records without a clear title prefix are labelled "Unattributed".
Comparison figures
The NAO figure of 74 deployed use cases comes from "Use of artificial intelligence in government" (National Audit Office, March 2024), based on an autumn 2023 survey of 87 government bodies with a 98% response rate. The Public Law Project's TAG register count of 55 is from PLP's written evidence (UAIG0024) to the Public Accounts Committee, January 2025. Both figures are necessarily lower bounds on the true scale of deployment.
About this page
This page is part of FutureOfAI.uk. It is updated automatically each time the page loads. If you spot a categorisation error or want to suggest an improvement, get in touch via the main site.