abc.net.au web signal

Australian government agencies miss AI transparency test

TL;DR

  • An ADM+S analysis identified 30 Australian government entities potentially in scope of the AI transparency policy that have no published statement at all.
  • The transparency standard became mandatory in February 2025 for 55 non-Corporate Commonwealth Entities and only required an overview of intended AI use.
  • A January 2026 OAIC report found only 13 of 23 assessed agencies mentioned automated decision-making in their Information Publication Scheme content.

The first round of Australia's AI self-reporting regime has not gone the way the Digital Transformation Agency probably hoped. As ABC News and other outlets covering the audit have reported, an analysis by the ARC Centre of Excellence for Automated Decision-Making and Society (ADM+S) identified 30 government entities that look to be in scope of the policy and have no AI transparency statement at all. Quality among the agencies that did publish varies widely, with some statements informative and others providing scant detail.

Some context on what the policy actually asks. The transparency standard became mandatory in February 2025 for 55 non-Corporate Commonwealth Entities, and at this stage it only required an overview of intended AI use. That is a low bar, which makes the non-compliance reading sharper: agencies are missing the easy test.

The Office of the Australian Information Commissioner has been running a parallel ruler over a related question, namely whether agencies disclose automated decision-making in their Information Publication Scheme content. In a January 2026 report, the OAIC found that of 23 agencies assessed, only 13 mentioned ADM at all, and only four (the ATO, Services Australia, the Department of Health, Disability and Aging, and the Department of Veterans' Affairs) explicitly disclosed using ADM in public-affecting decisions. Information commissioner Elizabeth Tydd said information about decision-making improves integrity, accountability and trust.

The honest caveat is that the reporting does not list which 30 entities ADM+S identified, does not say what enforcement the DTA will apply, and the OAIC sample was de-identified. The 13-of-23 disclosure rate and the 30-missing count are measuring different obligations, so they should not be added together.

What is worth watching is whether the regime gets teeth. If publication discipline tightens, with central listings, audits, and named agencies, then early adopters like the Australian Federal Police, which voluntarily published its statement, will look less like overachievers and more like the new floor.

Shared on Bluesky by 2 AI experts