Senior federal health officials discovered only days before a major aged-care funding tool went live that assessors would not legally be able to override its recommendations, despite having previously been told that human judgement could be used when the algorithm produced an inappropriate result.

Internal emails released under freedom-of-information laws show the issue triggered an urgent briefing for Aged Care Minister Sam Rae and Health Minister Mark Butler three days before the 1 November rollout of the Support at Home assessment algorithm. The documents reveal a sharp gap between how the system had been designed and described to assessors and what the Aged Care Rules 2025 actually allowed.

The Integrated Assessment Tool, or IAT, is used to collect information about an older person’s physical, social and personal circumstances. An algorithm then uses that information to recommend a Support at Home classification, which helps determine the level of government-funded assistance available to the person.

For months, the department had reassured assessment organisations that assessors would retain an ability to use clinical judgement and override the algorithm where necessary. An override function had even been built into the relevant systems.

But on 24 October, department assistant secretary Rachel Blackwood wrote to aged-care deputy secretary Greg Pugh after officials identified a legal problem. Her email said the Aged Care Rules had been drafted in a way that gave assessors and assessment delegates no discretion to depart from the algorithm’s recommended Support at Home classification.

That discovery meant the department’s previous guidance could not simply continue. If assessors were told to use an override that the rules did not permit, they could be making decisions outside the legal framework governing the program.

By 29 October, the problem had escalated to ministers. In the briefing prepared for Rae and Butler, senior staff warned that changing the message so close to the rollout carried its own risks. Assessment organisations had been prepared for a system in which professional judgement could correct a recommendation, and a last-minute instruction removing that option might not reach every assessor or be understood consistently.

The department also acknowledged that it had regularly told the public and stakeholders that an override would be available. That made the reversal more than a technical adjustment. It risked damaging confidence among assessors who had been told for months that they would remain responsible for applying professional judgement to individual cases.

The practical consequence was that the instruction manual for the IAT had to be changed before launch. Assessors would still conduct the face-to-face assessment and collect the information used by the system, but they could not manually replace the algorithm’s classification simply because they believed the recommended support level was too low or otherwise did not match the person’s needs.

The emails suggest the issue was not an obscure legal detail discovered after the fact. Blackwood told senior colleagues that many staff had regarded an override as a settled feature of the reform. She also warned that the department had been working to build credibility with assessment organisations and did not want to undermine those relationships at a critical point in the rollout.

The documents do not publicly reveal every option considered. Parts of the advice outlining possible solutions and risks were redacted. Pugh responded on 26 October that the approach selected might not have been Blackwood’s preferred option, but he considered it the best available path given the proximity of the 1 November start date and the need for a workable response.

It remains unclear from the released material why the department chose to change operational guidance rather than seek a legislative amendment before rollout. The Department of Health, Disability and Ageing did not respond to Guardian Australia’s request for comment on that question.

Concerns intensified after the system went live. Emails from November show assessment organisations began reporting cases in which they believed the algorithm was producing classifications that did not reflect clients’ needs.

On 20 November, Blackwood wrote that the majority of Aged Care Needs Assessment Organisations, including state and territory government bodies, had raised concerns about the inability to override the classification algorithm. She said organisations had provided examples where assessors believed the result did not align with the circumstances of the person being assessed.

Separate documents released under freedom-of-information laws show state health officials were keeping logs of hundreds of cases they considered under-assessments and sending them to federal counterparts. Those records reflect concerns raised by assessors and do not, by themselves, establish that every disputed classification was wrong. They do show that the inability to apply an immediate human override became a significant operational issue soon after launch.

Some assessors also raised professional and ethical concerns about being required to approve classifications they did not agree with. One email released with the documents warned that signing off on outcomes considered inconsistent with client needs could create professional and possible legal concerns for staff.

The controversy is especially important because Support at Home classifications are not abstract scores. They affect the amount and type of assistance an older person can access to remain at home, including support that may become more important as health, mobility or daily living needs change.

The documents also reveal the department introduced a workaround for some people who were already receiving support and were reassessed after their needs increased. Where the algorithm recommended a lower package, the portal could temporarily preserve the existing dollar amount even while a lower clinical classification was recorded in the person’s file.

That approach was intended to protect people from an immediate reduction in support, but it also highlights the difference between the clinical classification recorded by the system and the actual funding amount a person continued to receive. The government had publicly promised that older Australians would be “no worse off” as the reforms were introduced.

Opposition aged-care spokesperson Anne Ruston has argued that the law should be changed to restore a formal human override and says Coalition legislation to do that has passed the Senate but has not progressed through the House. Her criticism is part of the political response to the rollout and does not resolve the underlying administrative question of how the current rules should handle disputed algorithmic outcomes.

The wider issue is not whether automated tools should play any role in aged-care assessment, but how much authority they should have when experienced assessors believe an individual case does not fit the model’s recommendation. Algorithms can apply rules consistently across large numbers of cases, but consistency is not the same as accuracy in every individual situation.

The department’s own internal correspondence shows officials understood that tension before the rollout. They had built an override into the system, had communicated that it would exist and then discovered the legal rules did not support using it. That sequence is now central to questions about whether the implementation process adequately aligned policy, legislation, software design and frontline guidance before the system went live.

For older Australians and their families, the immediate concern is what happens when an assessor believes the algorithm has underestimated a person’s needs. For assessors, the issue is whether there is a clear, lawful pathway to escalate or correct a classification they consider inappropriate. For government, the challenge is to show that funding decisions remain accountable and clinically defensible even when the formal rules limit individual discretion.

The released emails provide no simple answer to those questions, but they do show that the lack of an override was not merely a later complaint from critics. It was a problem identified inside the department before launch, serious enough to trigger an emergency ministerial briefing and a last-minute change in instructions just days before the new aged-care system began operating.

The implementation problem also exposed how several layers of a reform can drift apart. Policy teams may intend for clinicians to retain discretion, software teams may build a button that allows an override, and training material may tell staff that the option exists. But if the governing rules do not authorise that discretion, the legal framework ultimately controls what assessors can do.

That is why the late discovery was so disruptive. It was not simply a software defect that could be patched overnight. Fixing it required either changing the legal settings, changing the operational guidance or creating another lawful pathway for disputed cases. The department chose to change the guidance immediately and manage the consequences after rollout.

The effectiveness of that approach will depend on whether the current system can identify questionable classifications quickly, protect clients from inappropriate reductions and give assessors a clear escalation route. The FOI documents show those questions were live inside the department before launch and became more urgent as complaints arrived after the system went into use.