Government agencies told to police own AI use missed first transparency test
Last year, the federal government opted for a softer model for regulating artificial intelligence. (Thinkstock)
In short:
Last year, Australia decided against an EU-style approach to regulating AI and instead tasked each government agency with managing their own use of the technology.
The ABC can reveal dozens of federal agencies missed mandatory deadlines to disclose how they are using AI.
With some agencies' transparency statements described as detailed and others scant, experts say it shows the challenges of the model adopted by the government.
Dozens of government bodies have failed the first test of policing their own use of AI, after Australia backed away from stricter European-style AI rules in favour of relying on existing regulators and agencies.
Last year, the federal government opted for a softer model for regulating artificial intelligence than the European-style approach of a single AI law with binding rules about high-risk AI.
The federal government said existing laws would suffice but tasked each agency with policing their own use of AI and, subsequently, regulating their own industries' use of the new technology.
But documents tabled in the Senate by the agency responsible for technology across government, the Digital Transformation Agency (DTA), showed the challenge of these approaches.
More than half of the government agencies checked by the DTA had not met the first transparency deadline for disclosing how they use artificial intelligence, according to documents tabled in parliament.
Fatima Payman said the records proved agencies were not even meeting their basic obligations. ( ABC News: Nick Haggarty)
Independent senator Fatima Payman, who pursued the issue through Senate estimates, said the records proved agencies were not even meeting their basic obligations.
"If the government can't even regulate its own AI use, how can Australians expect it to regulate AI in the private sector, which is reshaping the workplace for millions of Australians?" Senator Payman told the ABC.
DTA deputy CEO Lucy Poole acknowledged the challenge of regulating a fast-moving technology.
"The rate of change AI presents to the Australian Public Service means we need to work flexibly and considerately to ensure it is applied meaningfully and responsibly where appropriate," Ms Poole said.
Most government agencies missed AI disclosure deadline
Before the government's u-turn following the 2025 federal election, the DTA set a policy requiring government agencies to share some basic information about how they were using AI.
The DTA's use of AI in government policy was intended to build trust by being transparent.
Those transparency statements were meant to tell the public whether an agency was using AI, where it was being used, what safeguards applied, and whether people could be directly affected by automated systems.
But by 2pm on February 28, 2025 — a deadline set six months after the policy came into effect — only 40 of 92 mandated agencies had told the DTA they had done so, DTA records show.
Following the publication of this article, the DTA provided additional documentation showing that its subsequent compliance sweep of the agencies' websites confirmed only 13 were not compliant.
In addition, the DTA later told the Senate that 30 of 92 mandated agencies had failed to meet another new requirement to nominate someone from the agency who was accountable for the use of AI by its earlier deadline in November 2024.
Late, unaware and emails sent to spam
Some agencies were late. Others said they were unaware of the requirement or later corrected answers to Senate questions.
In one case, the Commonwealth Grants Commission initially gave evidence that it had met both deadlines before later acknowledging it hadn't.
Another agency, the National Competition Council, told a Senate estimates hearing it had not published its AI transparency statement by February 2025 because it was unaware of the deadline.
It later said it was compliant "by extension" because it relied on the ACCC for information technology services, and the ACCC had published its own statement on time.
The Organ and Tissue Authority (OTA) was initially treated as non-compliant by the DTA, but later correspondence showed the agency had emailed its accountable official details on December 5, 2024.
OTA's email had been caught in DTA's spam filter. The DTA later corrected the record and said that OTA had complied.
University of Sydney law professor Kim Weatherall said this first attempt at transparency statements had shown the challenges of assigning the responsibility to each of the agencies.
"It's just another big thing to add to their plate," Professor Weatherall said.
She said the first set of transparency disclosures were a "mixed bag".
Some were very detailed, she said, while others provided scant information about what types of AI they were using or what rules they had around its use.
But she cautioned against assuming a rough first go at transparency statements meant that government agencies weren't doing a good job of managing AI use within their departments.
"It's entirely possible they're doing good work internally while also dealing with a rapidly changing landscape,"Professor Weatherall said.
Government ramps up AI expectations
The DTA now says all 94 agencies have met their obligations around AI transparency and appointing an accountable official.
"As this was a new policy, the DTA worked closely with mandated agencies to ensure they understood their requirements and delivered on the expected outcomes," Ms Poole said.
It has also moved to make the AI transparency statements easier to find.
Previously, Australians had to search individual agency websites to compare statements about government AI use. Now, the DTA has created a central register of AI transparency statements.
"We will continue to iterate the register to improve the user experience and detail included," Ms Poole said.
DTA deputy CEO Lucy Poole acknowledged the challenge of regulating a fast-moving technology. (Supplied: DTA)
Government agencies have also been given new responsibilities for use of AI in government.
On top of the transparency statements, government agencies must develop a strategic position on AI adoption, keep internal registers of AI use cases, train staff in responsible AI use, and assess the potential impact of their use of AI.
The government's AI plan for the public service also expects every agency to appoint a chief AI officer in 2026.
It is a new role distinct from the accountable official covering the agency's compliance, and AI officers are tasked with driving adoption of AI within agencies.
Professor Weatherall said the framework still had a structural weakness because the DTA was not an enforcement body.
"They're trying to do the right thing — producing standards and guidance — but they're not a regulator," she said.
"They don't have enforcement powers."
Senator Payman said the agency compliance problems raised questions about the government's ability to regulate AI outside the public service.
"Numerous agencies have failed to meet basic AI transparency requirements, some of them months after they were required to," she said.
"There's no penalty for agencies who don't comply, so why would they?"
Finance Minister Katy Gallagher's office did not respond to a request for comment.