Feds move to regulate automated decision-making

The federal government will move to tighten regulation of automated decision making (ADM) as a result of a new set of safety priorities for adoption of AI in government and the community.

The Attorney-General has been tasked with leading the development of a framework to regulate the use of ADM within federal agencies.

The framework’s intent is to recognise “the importance of ensuring fair, accurate and transparent government decision-making”.

The new framework is among five AI safety priorities outlined by the government, including consumer protection from agentic retail price surveillance, better workplace safety, strengthening personal data protections and placing an onus on technology providers to design safety into their AI products.

However, ADM has become one of the most politically sensitive AI topics that the government has had to deal with in recent times.

This has in part been driven by the atmosphere left by Centrelink's disastrous algorithmic Robodebt debt collection program.

One of the 57 recommendations of the Royal Commission into the Robodebt scandal called for better legal and ethical governance of automate decision-making processes.

However, there has also been more recent controversy around the use of ADM across federal agencies, and concerns about its application in delivering social welfare services, including aged and disability care.

In April, the Office of the Commonwealth Ombudsman confirmed to several media outlets that it was conducting an investigation into complaints about the Department of Health, Disability and Ageing’s use of Integrated Assessment Tool (IAT).

Made mandatory in November last year as part of aged care reforms, the tool automated classification of recipients’ funding eligibility based on data collected by human assessors.

However, in many cases recipients who requested reassessments after experiencing a deterioration in their health found were classified for lower levels of support with assessors unable to override the tool.

A Senate inquiry has since heard that the department has received 834 requests for internal review of IAT decisions since it was introduced in November.

Since then, a coalition of senators including Anne Ruston, Penny Allman-Payne and David Pocock have passed a private members bill through the upper house to amend the Aged Care Act to ensure humans can override automated tools.

Specifically, it will make “clear that the Integrated Assessment Tool (IAT) and any other classification tools and requirements prescribed by the Aged Care Rules 2025 (the Rules) must operate to support, not replace, professional judgement of approved needs assessors”.

The bill is yet to pass the lower house.

Similarly, legislation to reform the National Disability Insurance Scheme (NDIS), which allows for more automation, has attracted criticism from human rights advocates.

The bill, the National Disability Insurance Scheme Amendment (Securing the NDIS For Future Generations) Bill 2026, will allow computer systems to make “evaluative determinations”.

According to the bill’s explanatory memorandum adds a new definition “evaluative determination” to mean “a discretion being exercised”, an “evaluative judgement [is] being made” or “a state of mind [is] being formed”.

In a LinkedIn post, Ed Santow, a former Human Rights Commissioner and co-director of the Human Technology Institute said that legislation would allow the relevant minister to automate decisions in cases where human judgements and evaluations would ordinarily be relied on.

He said that “this level of automation is particularly concerning where people's most basic rights are at stake. At the very least, the bill should provide much better safeguards against incorrect or unlawful decisions. Instead, it will be harder to review an automated NDIS decision”.

Automation disclosures

There has also been concern about the government’s openness about its use of ADM.

In January this year, the Office of the Australian Information Commissioner (OAIC) published research revealing a lack of transparency around the application of ADM across agencies authorised to use it.

It examined the websites of 23 agencies authorised to use the technology for compliance with the open government principles under the Information Publication Scheme (IPS) and found that only four (17 percent) disclosed that they used it.

A further nine made reference to ADM in their IPS disclosures but without confirming that they used it.

Those that did disclose that they used ADM “were not clear about how they used it”, the OAIC further found.

None of the agencies – reported to include the Australian Tax Office, Services Australia, the Clean Energy Regulator and the Department of Foreign Affairs and Trade – published guidelines or policies on how they use the technology.

That prompted the OAIC to recommend that all agencies authorised to use ADM disclose that status in their IPS, reveal what types they use, the decisions they apply it to – with examples – and any principles they apply to guide its use.

At the time Australia’s Information Commissioner Elizabeth Tydd said that the lack of transparency was placing public trust in government and ADM at risk.

"By clearly explaining when and how ADM is used, agencies can improve integrity, strengthen public confidence and uphold the statutory right to government information," Tydd said.