AUSTRALIA'S ROBODEBT SCHEME

In July 2015, Australia’s Department of Human Services approved an automated system to recover alleged welfare over-payments. The system compared Australian Taxation Office annual income data with Centrelink fortnightly welfare payment records. It calculated potential debts by dividing annual income by 26 fortnights, assuming stable employment and consistent earnings throughout the year.

This assumption was fundamentally flawed. Approximately 93% of welfare recipients work irregular, seasonal, or casual employment where income fluctuates significantly between fortnights. The crude averaging method systematically generated false debt notices against individuals who had received correct payments.

Between 2015 and 2019, the system issued approximately 470,000 incorrect debt notices totalling AUD 1.76 billion. Recipients experienced severe psychological distress, financial hardship, bankruptcy, and in some cases, suicide. The system reversed the onus of proof, requiring citizens to demonstrate they did not owe money rather than requiring the government to prove the debt existed.

In 2023, the Royal Commission into the Robodebt Scheme found that the scheme represented a ‘gross ethical failure’ and that senior officials had been aware of legal concerns but proceeded regardless.

This was not an abstract policy failure; it affected 470,000 people. The emotional weight of the case is essential to building conviction that governance matters.

It wasn't uniquely Australian nor governmental. The governance failures were universal: imprecise objectives, absent controls, and no contestability. Private sector companies must recognise these patterns in their own organisations.

Had the Ai Technical Standards been implemented, these would have been applicable. The failures compounded through the entire lifecycle:

• Design: Statement 10 (Human-centred Approach) — failure to consult welfare recipients or advocacy organisations

• Data: Statement 17 (Validate and Select Data) - No data validation of income-averaging methodology

• Evaluate: Statement 27 (Test for Specified Behaviours) — failure to stress-test against irregular employment patterns;

• Monitoring: Statement 38 (Ongoing Monitoring) — failure to capture and act on early complaints indicating systemic error.

The Australian federal government agreed in September 2025 to pay $475m in additional compensation to roughly 450,000 victims of the Robodebt scandal. The total financial redress to robodebt victims is now more than $2.4bn.