Australia’s aged care algorithm cannot legally be overridden, even when an assessor thinks its decision is clinically wrong, the Commonwealth Ombudsman has found. The report, released on 26 October, is about a government system, but its lessons apply to any Australian business that uses software or AI to make decisions about people, from insurance claims and credit checks to rostering, triage and hiring. Here is what the Ombudsman found and what to build into your own automated decisions so you don’t repeat it.
What the Ombudsman found about the aged care algorithm
The investigation, Automation and oversight in Aged Care, looked at the Integrated Assessment Tool (IAT), the digital assessment used for home support assessments since 1 November 2025. An assessor answers the tool’s questions about an older person, and an algorithm then makes automated decisions about their classification and priority, which determine the level of support they can get at home.
Commonwealth Ombudsman Iain Anderson formed the opinion that the Department of Health, Disability and Ageing’s administration of the system, including the IAT, had been “unreasonable”. The key findings:
- No override at law. The IAT appears to have been built with an override mechanism, but there is no legal mechanism to change automated decisions considered clinically inappropriate.
- A “technological black box”. Until recently there was no public information meaningfully explaining how the algorithm turns inputs into outcomes. The Department published its business rules and plain-English material on 6 October 2026, after responding to the draft report.
- No meaningful risk assessment. The report says the IAT was implemented in its current form “without any meaningful risk assessment, evaluation or analysis, despite the obvious risks” of automated decisions that do not allow human override.
- Wrong advice to users. Up until the legislation took effect, the Department had been telling assessment providers and the public that automated outcomes could be changed.
- Review that can only confirm. The report quotes a senior executive’s internal concern that “the review pathways only allow for the algorithm outcomes to be confirmed”, and finds the Robodebt Royal Commission’s recommendation on transparent automated decision-making has not been followed.
The Ombudsman is clear that the investigation did not assess whether the algorithm’s outcomes are clinically appropriate, only how it was designed, implemented and overseen. Those deficiencies, the report says, raise “serious questions” about its suitability and the need for independent clinical review.
How the government has responded
The report makes 10 recommendations. ABC News reports that six were accepted, while the first two, changing the law so assessors can substitute their own decision and updating the rules to match, were only noted. The government has instead legislated an escalation pathway for exceptional cases and commissioned a review of the assessment system. Anderson told Guardian Australia he is concerned the escalation pathway “will not necessarily” let assessors change decisions that appear clinically inappropriate. In a statement reported by SBS News, Ministers Mark Butler and Sam Rae said the government will consider the report carefully and is already acting on many of the recommendations.
Why the aged care algorithm matters beyond aged care
The IAT is a rules-based algorithm, not generative AI. That makes the findings more relevant to business, not less: if a deterministic system with published questions can end up as a black box with no way to correct it, an AI model that is harder to explain needs even more care. The same pattern turns up wherever software decides or recommends something that matters to a person, including claims triage in insurance, credit and hardship decisions in finance, care planning and intake in healthcare, and screening or rostering in HR.
There is also a legal deadline. From 10 December 2026, the OAIC says organisations covered by the Privacy Act must explain in their privacy policy where a computer program makes, or substantially and directly informs, decisions that could significantly affect people’s rights or interests. Our guide to the automated decision-making Privacy Act changes explains who is caught. For health tools there can be a further layer, as the TGA’s recent review of AI scribes showed.
Six aged care algorithm lessons for any automated decision
- Make the override real, not just technical. The IAT had an override button that the law did not allow anyone to use. Check that your policies, contracts and approvals let a qualified person change the outcome, and log who did it and why.
- Give review the power to change the answer. A review that can only re-run or confirm the result is not a review. Decide who can overturn a decision, how fast, and what happens to the person while they wait.
- Explain decisions in plain language. People affected, and your own staff, should be able to understand which information drove an outcome. If you cannot explain it, you cannot defend it, and the person cannot tell whether to challenge it.
- Re-test every version you run. ABC reports that testing was done on an original version of the algorithm built by a consultancy, and the version in use was modified with no external validation. Every change to rules, prompts or models needs its own testing before it goes live.
- Treat inputs as judgement and check the people giving them. The Ombudsman found it contradictory to rely on assessor judgement for the inputs but not for the outcome, and found weak checks on assessor training. Train the people who feed the system and audit their inputs.
- Use complaints and overrides as monitoring data. The report found complaints were not used to spot systemic problems. Track override rates, appeals and complaints, and investigate patterns rather than handling each case alone.
What this means if you are planning automated decisions
Automation and AI can make assessments faster and more consistent, but only if people can see how decisions are reached and step in when they are wrong. Build those controls in at the design stage, because adding them after launch is slower, more expensive and, as this report shows, sometimes blocked by the rules you have already set. Our custom AI development and AI automation work plans human review, reasons and audit logs in from the start. If your team manages people decisions, our free AI for HR course covers safe use, and what an AI audit includes explains how we review existing tools and processes.
This is general information based on the Commonwealth Ombudsman’s report, OAIC guidance and published media reporting as at 29 October 2026. It is not legal, clinical or compliance advice. Get advice on your own obligations before automating decisions that affect people’s rights, care, money or work.
Build automated decisions people can trust
An AI business audit maps where automated or AI-assisted decisions already happen in your business, where a human override and clear reasons are missing, and what the aged care algorithm findings mean for your own systems. Aideveloper builds decision tools with human review, plain-English reasons and audit logs planned in from the start. Talk to us about a review or a build.
Featured image: Photo by khbab alturky on Pexels.
