Snapshot checked at 2026-09-20 23:30:37 UTC. Counts can change.
Australia's central AI-transparency page listed 95 mandatory government-entity statements. The UK's algorithmic-transparency search returned 152 records. The US federal repository reported 3,611 individual AI use cases.
It is tempting to rank the countries from those totals. That would be wrong. Each number counts a different object.
| Comparison point | Australia | United States | United Kingdom |
|---|---|---|---|
| Disclosure unit | One high-level statement per covered entity | One individually reported AI use case | One published ATRS record describing an algorithmic tool in its stated use-case context |
| Coverage trigger | Covered non-corporate Commonwealth entities | Federal agency AI uses, subject to reporting rules | Collection: records published by UK public-sector organisations. Mandatory policy: covered central-government bodies using a tool that significantly influences a decision-making process with public effect or directly interacts with the general public |
| Lifecycle gate | Describes the agency's overall approach; reviewed at least annually and after material change | All development stages are included in the individual-use total | Mandatory publication applies at Beta/Pilot or Production; earlier-stage records may be submitted, and previously used tools remain recorded when retired |
| Material limits | Individual use cases are not required; corporate entities, Defence and the intelligence community are outside the mandatory policy | Some common commercial-tool uses are consolidated; national-security, intelligence, Department of War, some research and legally non-releasable material are excluded | The 152-record live collection is not a mandatory-scope denominator: it can include earlier-stage, retired and non-mandatory records; mandatory-scope and sensitive-information exemptions also apply |
| Valid inference | At least 95 covered entities had linked statements on the central page | The repository contains 3,611 individually reported cases; 1,818 were deployed or piloted | The live search exposed 152 published records within the ATRS collection |
| Invalid inference | Australia disclosed only 95 AI systems | The US had deployed 3,611 AI systems | The UK government used only 152 algorithmic tools |
Australia's standard is intentionally high-level. It requires agencies to describe why and how they use AI, identify broad usage patterns and public-impact categories, explain monitoring and protection measures, and provide an update date and contact. It explicitly says agencies do not have to list individual use cases. Its central page also says agencies—not the Digital Transformation Agency—are responsible for content, accuracy and currency.
The US repository has a much more granular unit. Its 3,611 figure covers individually reported use cases across all stages, not just live deployments. The same pinned repository reports 1,818 deployed or piloted uses and 445 high-impact uses. Even those figures remain bounded by reporting exclusions and a separate consolidated route for some common commercial products.
The UK's unit is different again. ATRS records describe algorithmic tools in context, including process integration, human review, method, phase, scale and other operational details. The live search count covers the published collection, while the mandatory rule is narrower by organisation, tool/use criterion and lifecycle phase. A smaller total can therefore coexist with deeper records, voluntary or earlier-stage submissions, retained retired records and selective mandatory coverage.
What this comparison adds—and what it does not
This article provides a dated, source-pinned comparison of disclosure unit, inclusion scope, lifecycle gate, exclusions and valid inference. It does not claim that variation in government AI disclosure is a new finding.
The OECD's Digital Government Outlook 2026 already compares national transparency mechanisms and describes implementation as underdeveloped; its earlier observation recorded 131 UK ATRS records. Pan et al.'s AITS-101 work examines variation in the coverage and structure of Australian transparency statements. Bano and Zowghi analyse 92 Australian statements and describe how structural compliance may still fail to serve higher-risk, lower-control stakeholders.
Five questions to ask before comparing an AI register
- Is the unit an organisation, a use case or a tool?
- Which lifecycle stages count?
- Is inclusion triggered by any AI use or only public impact?
- Which common products, sensitive activities or institutions are consolidated or excluded?
- Who validates the entry, and how current is the index?
Three-claim self-check
“The US has deployed 3,611 federal AI systems.”
Reject. 3,611 covers individual use cases across all stages; the repository separately reports 1,818 deployed or piloted.
“The UK government uses only 152 algorithmic tools.”
Reject. 152 is the live number of published ATRS records in a collection that includes voluntary, earlier-stage and retained retired records; the collection is not a census of every tool or a mandatory-scope denominator.
“Australia's 95 statements reveal 95 AI systems.”
Reject. The unit is a covered-entity statement, and individual-use disclosure is not mandatory.
The self-check is an explanatory device, not evidence that reader comprehension has been measured.
Transparent numbers are useful only after their unit and denominator are transparent too.
Sources and reproducibility
Australia: We counted linked list items between the page's exact section headings: 95 mandatory and 21 voluntary. The fetched page at the check time had SHA-256 b7432c392b62d29ea25de6a538e1e14e56e35de44d81cf2cb4bafd4c246cf3f1; the page stated it was last updated 23 June 2026.
Australia: central list of AI transparency statements
Australia: Standard for AI transparency statements
United States: The pinned README reports 3,611 individually reported use cases across all stages, 1,818 deployed or piloted, and 445 high-impact; its SHA-256 is c9cc8fc318466ba636c30f00e1f9a61e7088c782dbf8c3b573054070085a3cf9.
United States: 2025 Federal Agency AI Use Case Inventory
United States: inventory pinned at revision 06c7ebeef5b376524211042bd9673b2a7fefd3e3
United Kingdom: The unfiltered search returned 152 records at the check time. The fetched dynamic page then had SHA-256 ee654d34672ac6e9a8958df3468feab36043330527447bba4a83f82962fb2425; the timestamp, query and displayed result are the reproducible snapshot, not a claim that future page HTML will keep the same hash.
United Kingdom: live ATRS search
United Kingdom: mandatory scope and exemptions policy
United Kingdom: ATRS field standard
Prior work
OECD Digital Government Outlook 2026
Pan et al., The Creation and Analysis of Government AI Transparency Statements in Australia
Bano and Zowghi, AI Transparency: Governance Compliance or Stakeholder Requirements?
These are dated source totals, not measures of adoption, quality, compliance or completeness, and this article is not legal advice.