Debts by Averaging
Australia's Robodebt scheme raised welfare debts by averaging people's yearly income across the year, wrongly showing many as overpaid. Hundreds of thousands were chased for money they did not owe.
When an automated shortcut is wrong, who has to prove it: the government, or the person who opens the letter?
What happened
From about 2015 to 2019, the Australian government raised welfare debts automatically using a method called income averaging. It took a person’s yearly income from tax office records, spread it evenly across the year, and compared that average with the income the person had reported while receiving benefits. Anyone whose pay was irregular, such as casual workers, students, or people moving in and out of work, could look as if they had been overpaid when they had not been. Debt notices went out to hundreds of thousands of people.
Who spoke up
People who received the debts, and their families, pushed back through a public campaign under the name #notmydebt, through test cases, and through a class action. In 2019 a test case brought by Deanna Amato led the government to concede that debts raised this way were unlawful. At the Royal Commission that followed, people described what the debts had done to them, including a mother whose son died after being told he owed money.
What changed
In June 2021 the Federal Court approved a settlement of the class action, covering about 648,000 people: A$112 million in compensation including legal costs, about A$1.76 billion in debts withdrawn, and about A$751 million in refunds. The judge described it as a shameful chapter in the administration of social security.
In July 2023 the Royal Commission into the Robodebt Scheme, led by Catherine Holmes, delivered its report. It called the scheme “a crude and cruel mechanism, neither fair nor legal.” A second settlement, with A$475 million in compensation, was approved by the Federal Court in June 2026.
What to keep straight
- Headline figures for the first settlement vary, from A$1.2 billion to A$1.8 billion, because they combine cash, refunds, and cancelled debts. The cash compensation was A$112 million.
- There have been two settlements, in 2021 and 2026.
- The scheme was about welfare, not schools. It is here because it reached students, young people, and families in the same way it reached everyone else.
Talk about it
- People were asked to prove they had not been overpaid. Who should carry the burden of proof when a computer says you owe money?
- Averaging is a reasonable shortcut for some purposes. Why did it fail here?
- What would you need to see before trusting a letter that says you owe money?
Go further
The Human-in-the-Loop Economy
Why cultural context, embodied wisdom, and the capacity to sense what machines cannot may define the most valuable form of intelligence in the age of automation
Read the article Teach it Classroom Kits · Ages 8–12Bias in a Bag
Write a rule from a small sample, then test it on the whole bag. A concrete, physical first encounter with what happens when a rule learns from data that doesn't look like the group it's meant to describe.
View the kitSources
- Why Robodebt's use of income averaging lacked basic common sense (Peter Whiteford), The Conversation (16 March 2023).
- The Federal Court approves a $112 million settlement for the failures of the Robodebt system, Human Rights Law Centre (2021).
- Final report of the Royal Commission into the Robodebt Scheme, Prime Minister of Australia (7 July 2023).
- Government offers another $475m compensation for Robodebt victims, iTnews (4 September 2025).
- Australia's biggest class action settlement approved, AAP (June 2026).
- Grieving mother shares Robodebt experience with Royal Commission, SBS News.
More accounts to read
All Community Voices-
Families
A False Match, an Arrest at Home
Facial recognition wrongly matched a man's licence photo, and police arrested him in front of his daughters. His lawsuit changed how Detroit police may use the technology.
2020–2024 · Detroit, US Read the account -
Families
A Risk Score on the Hotline
A county scores child-welfare calls using public records. Research found racial gaps in its recommendations, and parents with disabilities believe it flagged them.
2016–2023 · Pennsylvania, US Read the account -
Students
The Summer an Algorithm Set the Grades
With exams cancelled, a model lowered 39 percent of teacher-assessed A-level grades, and small classes fared best. Students marched; the government reversed within days.
2020 · England Read the account
Summarised from the public record. Follow the sources for the full story.