Since February 2025, every Commonwealth agency has had to publish an AI transparency statement: a public account of how it uses artificial intelligence, refreshed at least once a year. The APS AI Transparency Tracker reads all of them so you don’t have to. It collects every statement once a day, records each change in version control, and publishes the result as a live public register of how government describes its own use of AI.
Marking the government’s homework#
A transparency statement is only as good as your ability to check it. Published today, it can be softened tomorrow, and once the page is overwritten the original is gone. Across more than a hundred agencies, each on its own website and its own schedule, the record is almost impossible to keep an eye on by hand.
The tracker keeps it instead. Each day it saves a fresh copy of every statement and commits it to Git, the version-control system developers use to log every edit to their code. Nothing is silently overwritten. A timeline shows every change to every statement, newest first, down to the words that moved.
The template effect#
The most revealing view sets the agencies against one another. The Digital Transformation Agency publishes a model statement, and many agencies adopt it almost wholesale, changing a name here and a system there. The tracker measures how much of each statement is an agency’s own writing and how much is inherited boilerplate, then ranks them. A few have written something genuinely their own. Others have published the template with barely a word altered, nothing bespoke at all.
The site offers three ways into that pattern. A colour-coded wall sorts every agency into bespoke, borrowed, not-yet-published or exempt. A similarity map places statements that read alike near one another, warm where the wording is templated and cool where an agency wrote its own. A propagation view follows single passages as they spread from the template across dozens of statements.
Why the Studio cares#
Much of the Cybernetic Studio’s work is about making computation legible: slowing it down, blowing it up, bringing it within human reach. The tracker turns that same attention on the institutions that run the computation. How a government uses AI is, in the end, a question answered by the paperwork it files about itself, and paperwork is easiest to trust when someone is quietly keeping every draft.
A Cybernetic Studio project by Ben Swift. The register is live at anucybernetics.github.io/aps-ai-transparency-tracker, and the statements, with the code that gathers them, are public on GitHub.
