Social Security Appeals and Access to Justice: Learning from the robodebt controversy

This article – ‘Social security appeals and access to justice: Learning from the robodebt controversy’ – first appeared in Precedent, the journal of the Australian Lawyers Alliance, issue 158, published in June 2020 (Sydney, Australia, ISSN 1449-7719). It has been reproduced with the kind permission of the author and the ALA. For more information about the ALA, please go to:

The consent orders made in the recent Federal Court matter of Amato v the Commonwealth of Australia[i] (the Consent Orders) have triggered a refund of all averaged debts created under the government’s recent employment income confirmation initiative.

How did a flawed interpretation of the Commonwealth’s debt-raising powers take hold and persist through four years of criticism? And what does the affair tell us about how internal and tribunal appeal mechanisms are functioning? While public discussion of the robodebt[ii] controversy has often focused on technology or algorithms, this article argues that it also represents a familiar tale of what occurs when we fail to promote access to justice. In an era where technology can facilitate the making of hundreds of thousands of decisions, the effective, prompt governance of legal interpretation has never been more important.


The Amato litigation has succeeded in defining the specific cohort of unlawful approaches which are colloquially known as ‘robodebts’. The relevant legislative provisions, ss1222A and 1223(1) of the Social Security Act 1991 (Cth), position the Commonwealth as the entity which asserts and proves the existence of the debt. Their structure also reflects the long-established principle that in administrative decisions, the person who asserts the existence of a state of affairs must adduce adequate supporting material to establish that position.[iii]

Amato supports the principle that a calculation which uses averaged data matched from the person’s pay as you go (PAYG) summary does not meet the standard of certainty required by the statutory provisions. The social security law at the time required that a person’s entitlements, and in turn their debt, be calculated using fortnightly earned (not received) gross income.[iv] The Department of Human Services’ (the Department) now discontinued approach was to data match with annual PAYG statement data, and then divide it to obtain an average fortnightly figure. This averaging populated a crude assumption: that the person earned evenly across each fortnight, even where the person’s past income reporting showed variation. Fundamentally, therefore the apportioned data was not accurate or probative for people like Ms Amato: those with multiple employers, young people with varying hours and others in casual employment. The data was not capable of reflecting the variations in loading or entitlements that increasingly occur in the modern workplace. The Consent Orders thus refer to the decision to issue Ms Amato’s debt as ‘irrational in the requisite legal sense’.[v]

Significantly, the Amato litigation was a challenge to the later version of the income compliance program. In contrast to the original online compliance initiative, the later version featured human actors  at the point of decision – staff would apply the Australian Tax Office (ATO) data match after individuals failed to provide the requested historical payslip information. The litigation successfully uncovered a key direction to staff in the relevant operational blueprint which stated that while ATO data could be ‘suitable’ for the purposes of issuing a debt, the individual’s original reporting, without further evidence being supplied, was not.[vi] While the public imagination around robodebt is dominated by images of ‘algorithms’ or incorrect code, a fundamental flaw of the program was an unwillingness to recognise that an average does not speak logically to its parts. Where an individual had a track record of reporting in a particular way, this needed to be properly valued, not disregarded without clear justification. The Consent Orders therefore specifically noted that there was probative material before the decision-maker indicating that Ms Amato had not earned the same amount each fortnight.

The consequence of the Amato litigation is a concession by the government that a debt calculated solely from averaged data, in whole or in part, should be set aside. One week before the Consent Orders were made, Services Australia announced that it would cease using the methodology and would secure further proof points going forward. It also announced that it would begin the process of identifying and reviewing averaged debts that had already been raised. For legal professionals seeking to assist the cohort in the coming years, it is important to note that the averaging flaw applies only to payments calculated by reference to individuals’ fortnightly earnings.[vii] In order to have received a robodebt, the person must have been working for at least part of the period for which the debt was issued. The averaged debts will usually, but not always, have been issued after July 2015.[viii] They generally relate to earnings and reporting from 2009 onwards.

These date ranges reflect the fact that prior to July 2015, it was standard procedure for the Department to exercise its information gathering powers under s192 of the Social Security Administration Act 1999 (Cth); it would directly obtain relevant evidence from banks and employers. From July 2015, the Department placed the onus of obtaining payslip or other financial information onto the welfare recipient. Where the information was unobtainable or not supplied, the Department’s initial letters asserted the right to average the debt. This shift in evidential practice allowed the average compliance caseload to shift from 20,000 files per year to over one million initial compliance reviews in three years. Of the estimated 650,000 employment income debts, only one-third of the people who received the request for documents completed that process. While around 110,000 people later had their debt reassessed based on evidence they supplied, 470,000 are constituted by averaged data, in whole or in part.

A class action has now been launched by Gordon Legal seeking equitable and tortious remedies, with the aim of securing refunds, interest and compensation. The central claim of unjust enrichment effectively seeks to systematically apply the logic of the Amato litigation by arguing that averaged debts constituted unlawful demands for payment. Over $721 million of repayments will be returned, with the total value of the debts being greater than $1 billion.


The balance of this article considers how the robodebt controversy should inform further scrutiny of our system of social security appeals. As the peak body of the appeals process, it is important to explore and learn from the dynamics which limited the impact of the Administrative Appeals Tribunal (AAT) in these cases.

The role of the AAT in the appeals system is often presented through a deficit lens – it is simply not a court. Of course, as a de novo merits appeal body, the AAT is standing in the shoes of the original decision-maker and often dealing with changed evidence. Nevertheless, the statutory mandate of identifying the ‘correct or preferable’ answer requires AAT members to adopt a view of the law in every matter before them. While the final outcome will most often involve determining the ‘preferable’ discretionary factors, the AAT must, when necessary, rule out legally incorrect approaches in its reasoning.

On the final day of the first senate inquiry in May 2017,[ix] senators questioned the Department about a March 2017 AAT tier 1 decision, which had been provided to it by Victoria Legal Aid. This decision found that:

‘A fortnightly income test applies. In this case, no effort has been made by Centrelink to obtain actual wage records … even though such records would very likely be readily available if required. Instead it has simply been assumed that the total year earnings can be apportioned equally to each fortnight across the relevant financial year. However, that is not consistent with the requirements of the legislation. The actual pay records are critical to the proper calculation of the overpayment. Accordingly, Centrelink will need to request and obtain those records from the employer in order to arrive at a correct debt calculation.’[x]

The Department’s response to this decision was that it was ‘an AAT tier 1 decision, obviously turning on particular facts of that particular matter’.[xi] These dynamics continued when Prof Terry Carney, a long-serving member of the AAT, publicly advocated against the scheme, and disclosed that in 2017 he had made five AAT decisions challenging the approach taken to the legislation. These decisions are now publicly available, having been released to the Senate Standing Committees on Legal and Constitutional Affairs with the applicants’ personal information redacted.[xii] In the author’s view, the decisions cogently argue that the use of ATO data is ‘incorrect’ and not legally appropriate. In March 2020, Services Australia repeated its position that the decisions needed to be approached on the basis that they were tied to their particular circumstances.[xiii]

These exchanges embody the need for greater recognition of the specific role given to the AAT in overseeing the social security law. Section 8(f) of the Social Security (Administration) Act 1999 (Cth) requires that in administering the law, the Secretary is to have regard to ‘the need to apply government policy in accordance with the law and with due regard to relevant decisions’ of the AAT.

The specific inclusion of s8(f) in the schema of administration was a recognition of the systemic importance of AAT decisions for disadvantaged and vulnerable welfare recipients. It was a recognition of the Tribunal’s role in governing and co-ordinating the interpretation of the social security law.

The robodebt affair underlines the importance of resourcing the AAT so that it has a greater capacity to play a guidance role in particular areas of its jurisdiction. This reform would also recognise the unprecedented caseload pressure caused by higher volumes of technology-enabled decisions. Our appeals system needs to respond to a new reality, where over 150,000 debt decisions can be taken annually in one division of the Department. In learning from robodebt, we must work to ensure that a balanced caseload of appeals, with appropriate legal professional support, reaches the AAT. The Tribunal should be provided with specific powers allowing for accelerated case handling techniques and escalation procedures. These would enable it to more effectively prioritise areas of jurisdiction which have been the subject of major legislative or administrative reforms.

The well-documented challenges the AAT has faced in its NDIS jurisdiction also supports this need for greater interpretive focus and resourcing. In an era where technology is driving higher numbers of decisions, statutory interpretations underpinning system business rules must be stress tested as early as possible. While determining outcomes for individuals must always be its focus, the AAT also functions as an early warning mechanism for interpretive issues. A more proactive governance of interpretation is key to ensuring that flaws do not grow to the scale of robodebt. Services Australia should also be placed under obligations, similar to the Australian Taxation Office, to publicly declare its position on decisions of significance.  It should also desist from laying down evidential and interpretive approaches in internal operational blueprints – not currently published on its disclosure log – rather than publicly in the Guide to Social Security Law.

Despite this, however, it cannot be assumed that all meritorious cases are appealed – the fact that the Department’s belief in averaging as a lawful method appears to date back to the 1980s underlines this.[xiv] For reasons explored below, only a tiny proportion of robodebts ever exited the internal review process to the AAT. These numbers include debts for which documentary evidence had been tendered, as well as averaged debts. From June 2016 to August 2019, only 1,482 debts from the program were reviewed by AAT tier 1.[xv] The variation or set aside rate at AAT tier 1 was 42 per cent. Of the 174 debts from the program that proceeded to tier 2, 92 per cent were settled by consent. In its evidence to the second senate inquiry,[xvi] one month before the Amato decision, the Department argued that the appeal statistics supported its view that 99.2 per cent of the issued debts were accurate. This illustrates that the internal appeal processes are the central mechanism for most welfare recipients, who often lack the capacity and knowledge to persist all the way to the AAT.


The robodebt experience underlines that the AAT is a peak body that is dependent upon an accessible internal appeals system. The likelihood that an AAT complaint will be made is shaped by individuals’ resources, their networks and their legal consciousness. The existence of thousands of averaged debts at the time of writing underlines that many legal disputes are not recognised as such and it cannot be assumed that an aggrieved person will act. This represents a powerful endorsement of the findings of the NSW Law and Justice Foundation study of legal needs.[xvii]

In this last section, I will consider what options individuals had to obtain outcomes under the available internal review mechanisms.

The first action many individuals could take was to challenge the coercive and uncertain nature of the initial letters sent to them. These letters asked individuals to update their details from seven years ago and alluded to the possible issue of a debt notice if pay records were not supplied. These letters were usually framed as informal invitations rather than formal notices marking a statutory ‘event’. They proved effective as a behavioural ‘nudge’, with many Australians going to extraordinary lengths to secure payslip information on behalf of the Department.

The letters did not state whether there were any circumstances in which the Department was willing to go directly to employers to gather pay records. Following the first Ombudsman report,[xviii] the Department committed to introducing an ‘exceptional circumstances’ policy where it would use its s192 information gathering power. This policy remained unpublished until the Ombudsman’s follow-up review 18 months later,[xix] when it was finally incorporated in the Guide to Social Security Law.[xx] At the second senate inquiry, it emerged that Services Australia had only issued s192 notices to employers in 1,000 cases.[xxi] It was unable to answer whether those were instances where it chose to gather the records or where it was implementing an AAT order for the debt to be recalculated.

This data underlines the inequality of arms and asymmetry of information which shaped the frontline interactions and decisions. The senate has been consistently provided with instances where individuals have been unable to secure evidence of earnings due to their lack of resources (for example, to pay for historical bank statements) or because the businesses which they previously worked for have since closed.[xxii] The Amato litigationhas, however, fundamentally reset this position. With averaging now off the table, debts must be calculated using pay records or other probative evidence of the person’s earned income. Individuals should now be able to ask Services Australia to obtain these records. The government has not as yet, however, made a commitment to proactively gather the evidence if undertaking any recalculation.

There is also limited community awareness about the complexities involved in cases where individuals chose to tender bank statements as evidence of earnings. While a bank statement discloses the amount and frequency of money received by the person, it can be difficult to reverse engineer their gross earned income. The person’s fortnightly gross earnings must also be calculated according to the Centrelink calendar, rather than the payroll calendar. Where bank statements are used, it is also necessary to tease out any assumptions that were made about an individual’s tax position, working credits or deductions in the absence of pay records. As is reflected by the single touch payroll legislation, it is the contemporaneous pay records which provide the most reliable and efficient basis for any debt decision. Serious legal questions accompany the use of the bank statements by the agency-  the statute requires a calculation based on a confirmed facts, not a generic estimate.

Once a debt is issued, every person had a statutory right to lodge an application for a review by an authorised review officer.[xxiii] Even post-Amato, there is significant benefit to taking this step – it ensures that any process of review is carried out by a senior member of staff. If all people within the scope of the Amato decision had immediately taken this option, it would have represented an unprecedented workflow challenge. Throughout the program’s operation, 49 per cent of the debts reviewed by an authorised review officer were varied.[xxiv] Only 7,421 debts from the program – 1 per cent – ever were reviewed in this way.

The process of navigating and achieving an outcome using the internal appeals system too often depended on the person’s tenacity in interacting with multiple people across the different levels of the organisation. When the debt was first issued, individuals were offered an ‘informal’ check on the debt. Throughout the process, they were continually asked to provide new information. In these interactions, averaging was continually presented as a wholly acceptable option. In order to alter the original debt, the individual was urged to change the information upon which it was calculated. There have been over 110,000 successful reassessments since the program’s launch, which underlines the crude assumptions underpinning the ATO data. Significantly, the average lengths of time it took to obtain an informal ‘check’ or have debts reassessed were not recorded. These unacceptable gaps in the data which the agency could report on, embody how the selectivity of evidence base upon which the programme proceeded.

Many individuals were unaware of the difference that a few simple phrases such as ‘I want a review not a reassessment’ could make. For the past four years, ordinary Australians have devoted hundreds of hours to unravelling and correcting the crude assumptions that populated the initial letters. Even where documents were eventually obtained, the asserted right to average shaped all interactions around a fundamental power imbalance. It asserted that any silence or inaction on the part of the individual would be resolved against the person. The robodebt affair risks generating a profound sense of distrust in government advice among those affected. Not least of all because, up until February 2017, the Department’s own website advised recipients that they only needed to retain payslips for six months.

The existence of so many averaged debts shows that many viewed the debt as incontestable or the process as insurmountable, and simply gave up. The reverse onus structure always risked the debt amounts reflecting the resources, vulnerabilities and misunderstandings of citizens rather than proper documentary evidence. Given this, it is disappointing that internal systems within Services Australia were not set up to record which debts were averaged. In order to identify all debts which were partially or totally calculated using ATO data, 649 staff have now had to manually check 594,636 debt records. The final figure will powerfully embody the real-life impacts of the administrative burden. It will testify to the complex forms of triage and self-care that have come to shape vulnerable or disadvantaged people’s engagement with our social security system.


This article argues for two streams of reform to social welfare appeals following the robodebt controversy. First, we must properly value and strengthen the AAT’s ‘correct answer’ jurisdiction to ensure that legal issues can be tested and resolved quickly. While merits review tries to identify the ‘preferable’ exercise of a discretion, it is also a key forum for the early interrogation of the limits of legal authority. It was also always intended that institutions such as the Ombudsman and the AAT would refer matters to the Federal Court where appropriate. Government and oversight institutions should fund and facilitate the testing of evidential approaches and legal interpretations. The robodebt affair underlines how relatively minor investments in legal aid bodies and community legal centres can prevent the compounding of profound financial and legal risks.

The second stream of post-robodebt reform must centre on promoting greater legal consciousness among, and practical support for, social security recipients. The current class action by Gordon Legal will explore what systemic remedy is available to those who were issued with unlawful debt notices. While a full substantive analysis of that action is beyond this article’s scope, it will ignite a much-needed debate concerning the oversight and administration of the social security system. We have built a dense, highly conditional statute book which concentrates enormous powers to interpret, apply and determine important matters in Services Australia. The work of Victoria Legal Aid in framing the Amato litigation is a standout example of the profound public benefits that can flow from the presence of specialist social security lawyers. The enormous demand and funding pressures on community legal centres who are members of the newly launched peak organisation, Economic Justice Australia, must be tackled.[xxv]

The fallout from the robodebt affair is an opportunity to advocate for a specific charter of rights for our social security system. It reminds us that the content of our casebooks is too often shaped by socio-economic disadvantage and the absence of legal representation.

[i] Federal Court of Australia, General Division, Consent Orders of Justice Davies, 27 November 2019, File No. VID611/2019 (Consent Orders).

[ii] The author uses the term ‘robodebt’ to describe the specific cohort of unlawful debts which were issued reflexively using averaged data matched from the Australian Taxation Office (ATO). This also reflects the recent Macquarie Dictionary definition. Unlike the common public usage, the article does not refer to the employment income confirmation program as ‘robodebt’. Reflecting the term’s origins on social media, the usage of ‘robodebt’ rather than robo-debt is to be preferred. The author tabled a document defending the appropriateness of the term at the second senate inquiry: Senate Community Affairs References Committee, Parliament of Australia, Centrelink Compliance Programme, Melbourne, 9 October 2019, 14 (Dr Darren O’Donovan).

[iii] See further Power v Comcare [2015] FCA 1502; Re Martin and Commonwealth (1983) 5 ALD 277, 287.

[iv] With the rollout of Single Touch Payroll and the passage of the Social Services and Other Legislation Amendment (Simplifying Income Reporting and Other Measures) Act 2020, the test will shift to one of received income. The commencement of this legislation has been temporarily suspended until after the current coronavirus crisis.

[v] Consent Orders, above note 1, para 9.

[vi] Federal Court of Australia, General Division, Amato v the Commonwealth of Australia, Statement of Agreed Facts, File No. VID611/2019, paras 14.4 and 14.5.

[vii] These are: Newstart Allowance (47 per cent of all debts in the program); Youth Allowance (26 per cent); Disability Support Pension; Austudy Allowance; Age Pension (a very small proportion generated in 2016–17 only); Carer Allowance; Parenting Payment; Partner Allowance; Sickness Allowance; Special Benefit; Widow A Allowance; and Widow B Pension.

[viii] Foonote 101 of the Commonwealth Ombudsman’s first report into the program indicates that the Department has held the view that averaging was permitted since the 1980s. While the ‘robodebt’ controversy entered the media in late 2016, the reverse-onus approach commenced via the ‘interim manual’ system on July 2015: Office of the Commonwealth Ombudsman, Centrelink’s Automated Debt Raising and Recovery System, Canberra, April 2017 (Ombudsman’s first report), paras 1.13–1.16.

[ix] Senate Community Affairs References Committee, Parliament of Australia, Inquiry into the Better Management of the Social Welfare System Initiative, 2017.

[x] Decision of Member Treble, Administrative Appeal Tribunal, 2016/M103550 (24 March 2017), para 17.

[xi] Senate Community Affairs References Committee, Parliament of Australia, Thursday 18 May 2017, Inquiry into the Better Management of the Social Welfare System Initiative, Official Hansard, 47.

[xii] Senate Standing Committee on Legal and Constitutional Affairs, Parliament of Australia, Supplementary Budget Estimates 2019–20, Answer to Question on Notice LCC-SBE19-2.

[xiii] Community Affairs Legislation Committee, Parliament of Australia, Thursday 5 March 2020, Additional Budget Estimates, Official Hansard, 140.

[xiv] Ombudsman’s first report, above note 8, footnote 101.

[xv] Senate Community Affairs Legislation Committee, Parliament of Australia, Answer to Question on Notice 131 (SQ19-000240), 12 December 2019.

[xvi] Senate Community Affairs References Committee, Parliament of Australia, Centrelink Compliance Programme, Canberra, 9 October 2019 (Services Australia Representative), 47.

[xvii] Law and Justice Foundation of New South Wales, Legal Australia-Wide (LAW) Survey,Sydney, 2012.

[xviii] Ombudsman’s first report, above note 8, Recommendation 4, 28.

[xix] Office of the Commonwealth Ombudsman, Centrelink’s Automated Debt Raising and Recovery System—Implementation Report, Canberra, April 2019, para 2.58.

[xx] Department of Social Services (Cth), Guide to Social Security Law, ch 6.3.9,

[xxi] Services Australia (Cth), Response to Community Affairs References Committee, Parliament of Australia, Question on Notice, QoN 44, 7 February 2020.

[xxii] See eg, the opening statements of two Victoria Legal Aid clients, later republished: see L Luty and K O’Shea, ‘The system is truly Orwellian’: Our horrifying experiences with robodebt’, The Guardian (online), 10 October 2019, <>.

[xxiii] Social Security (Administration) Act 1999 (Cth), s129.

[xxiv] Services Australia (Cth), Submission No. 20 to the Senate Community Affairs References Committee, Inquiry into the Income Compliance Programme, 23.

[xxv] In early 2020, the National Social Security Rights Network was relaunched as Economic Justice Australia. For more information, see <>.

Darren O’Donovan