State Health Officials Flagged Aged-Care Algorithm Failures Within Days of Launch
State health officials kept logs of hundreds of aged-care cases where older Australians were severely under-assessed by the federal government's algorithmic in-home aged care funding tool, according to documents obtained by Guardian Australia under freedom of information laws. Officials from four states and the Northern Territory recorded systemic errors in the assessment system since the Integrated Assessment Tool (IAT) was rolled out the previous November. Guardian Australia
Here is how the tool works. The IAT uses a questionnaire completed by aged care assessors from state and territory health departments. The answers are fed into an algorithm — a set of automated rules — that determines the level of funding a person receives for in-home support packages and how urgently they need help. The algorithm was designed by federal health and ageing staff led by four assistant secretaries overseeing five departmental data scientists, a principal cost modeller and a director. Days before the IAT was rolled out, the federal health department barred assessors from overriding the tool's decisions.
Within days of the tool going live, people were being under-assessed by the algorithm, according to documents released by the federal Department of Health and Aged Care. Assessors reported feeling powerless and ethically challenged. Clients were left at risk of deteriorating, remaining in hospital, or being forced into aged care.
WA Health director general Dr Shirley Bowen emailed her federal counterpart on 24 November warning that without a pathway for discretion, a small but vulnerable group of older people might experience "catastrophic outcomes" while waiting for algorithmically determined access to services. Bowen's email also warned the algorithm created risks of further discharge delay and blocking access to acute care for others who need it.
The warnings came early and from multiple jurisdictions. On 7 November, a staff member with Aged Care Assessment Victoria emailed the federal health department about concerning outcomes from the IAT algorithm and the inability of assessors to override the result, saying assessors felt conflicted as they did not agree the IAT outcome reflected clients' needs. On 13 November, a representative from NT Health emailed the federal government about concerns regarding the IAT algorithm and the erosion of clinical judgement within the new assessment system, attaching examples of clinical assessments with unexpected outcomes and saying assessors and delegates were disheartened they could no longer use their clinical judgement and experience.
Health Minister Anika Wells announced the IAT as a new tool to improve aged care assessments when it was first unveiled, describing it as the new tool for assessing the eligibility of older people for government-subsidised aged care. health.gov.au The department's own user guide, updated as recently as 28 August 2026, says the IAT aims to drive good practices in the assessment of older people's support needs and eligibility for Australian Government-subsidised aged care. health.gov.au
The FOI documents add to a pattern of reporting on the tool's failures. In July 2026, Guardian Australia reported that the tool frequently under-assessed people and left them without adequate care, and that Aged Care Minister Sam Rae was defending the algorithm. Guardian Australia In August, the Greens said the IAT was under-assessing far more older Australians, many of whom may not know how to contest the outcome, and that a fifth of older Australians who contested an automated decision had it changed. Guardian Australia
What the FOI documents make clear is that the problems were not discovered months later. They were flagged within days of the tool going live, by the very assessors the system depends on, and escalated to the most senior levels of state health bureaucracies within weeks. The department was told, in writing, that its algorithm was producing results assessors considered wrong, that the override ban was removing the safety net, and that the consequences for a vulnerable population could be catastrophic. Those warnings were specific: hospital discharge delays, blocked acute care beds, people forced into residential aged care who could have stayed at home.
The design choices here matter. The algorithm was built by a relatively small team within the federal department, and the decision to bar assessors from overriding its outputs was made days before launch, not after a trial period. That is a governance choice. It means the department prioritised algorithmic consistency over clinical discretion, and did so before it had any real-world evidence that the algorithm was producing sound results. The emails from Victoria and the NT, sent within the first fortnight of operation, suggest it was not.
A one-in-five reversal rate on review, as the Greens highlighted in August, is the kind of figure that would trigger a serious audit in most public-sector algorithmic decision-making systems. Think of it this way: if one in five people who appeal a parking fine get it overturned, you might question whether the ticketing system is working properly. The FOI documents now show that the people closest to the tool, the assessors running it day to day, were raising the alarm from the very first week. The question for the department, and for Minister Rae, is not whether the tool has problems. That is established. It is what has been done about them in the months since those first warnings landed.
