Content uploaded by Andrew Ray
Author content
All content in this area was uploaded by Andrew Ray on Jan 03, 2021
Content may be subject to copyright.
1
Committee Secretary
Senate Standing Committees on Community Affairs
PO Box 6100
Parliament House
Canberra ACT 2600
16 September 2020
Dear Committee Secretary,
RE: Inquiry into Centrelink's compliance program
The Australian National University Law Reform and Social Justice Research Hub (ANU LRSJ
Research Hub) welcomes the opportunity to provide this submission to the Senate Standing
Committee on Community Affairs regarding Centrelink’s Compliance Program.
The ANU LRSJ Research Hub falls within the ANU College of Law’s Law Reform and Social
Justice program, which supports the integration of law reform and principles of social justice into
teaching, research and study across the College. Members of the group are students of the
ANU College of Law, who are engaged with a range of projects with the aim of exploring the
law’s complex role in society, and the part that lawyers play in using and improving law to
promote both social justice and social stability. The group has also collaborated with a student
from the ANU College of Engineering & Computer Science.
Summary of Recommendations:
1. The Committee should delay its final report so that it can consider any comments made
by the AHRC regarding the Compliance Program, and future best-practice.
2. The Committee should call on Government to agree to repay interest to individuals who
paid money in response to unlawful debt notices.
3. The Committee should consider if, as a matter of policy, the government should waive all
debts arising from the operation of the compliance program to ensure that future
compliance activities are not impacted by a reduced willingness to cooperate and supply
requested information. This should include debts raised on the basis of information
supplied to Centrelink in response to an initial notice.
4. Any information gathering powers provided by s 66A should be clarified, and the section
should be amended to expressly include the protections afforded in ss 192-198 (regarding
the required standard of the notice).
5. That Services Australia commit to a more efficient rollout of debt repayments, and that a
greater effort is made to explain what resources and programs have been committed to
this effort.
2
6. That Services Australia should be obligated to consider the wellbeing impacts on
individuals, particularly those most vulnerable, when making decisions for social security
payments.
7. A review of algorithmic decision-making’s appropriateness in the specific context of debt-
raising should precede its resumption in any form.
8. Humans should maintain involvement in decision-making where those decisions will have
a significant impact on individuals.
9. The Freedom of Information Act should be amended to make it easier for individuals (and
the Senate) to access information relating to the use of automated systems by
government.
10. The human costs from the Compliance Program’s use should motivate a strengthened
commitment to government support and consultation of both technical research into
algorithmic explainability/interpretability, and legal research into effective algorithmic
regulation.
11. This research should be prioritised commensurate to the government’s intent to deploy
and rely on automated decision-making systems.
12. The government should prioritise the creation of generally applicable standards of
automated decision-making, taking in the rapid advances of AI and machine-learning
technology.
13. Future compliance activities adopt a principles-based approach to debt recovery.
14. Any further review of the compliance program should consider the complexity in the Acts
underpinning its operation and amend those Acts to improve the openness and
transparency of decision-making practices.
If further information is required, please contact us at anulrsjresearchhub@gmail.com.
On behalf of the ANU LRSJ Research Hub,
Authors: Andrew Ray, Annika Reynolds, Charles Evans, Pippi Callan, Sathyajith Sukumaran,
Sebastian Judge
Editor: Jessica Hodgson
3
Introduction
The Inquiry into Centrelink’s Compliance Program was referred to the Committee on 31 July 2019.
In response to a series of ‘significant developments’, the Committee reopened submissions to
address the original terms of reference with particular interest regarding:
● the legal basis of the online compliance program;
● Services Australia's progress in implementing the changes announced in November 2019;
● the impact that these changes have had on individuals, the community sector and
Centrelink staff; and
● the future of Centrelink compliance activities and programs, particularly compliance
activities for the recovery of social security overpayments.
This submission will focus on the supplementary terms of reference and discuss a broader point
regarding the legal basis for, and best practices regarding the use of, automated decision-making
by government. We note that automated decision-making and AI-assisted decision-making is
presently the subject of an AHRC discussion paper
1
and recommend that the Committee delay
its final report until it can consider any comments made by the AHRC in its ongoing review of
automated decision-making.
Recommendation 1: The Committee should delay its final report so that it can
consider any comments made by the AHRC regarding the Compliance Program, and
future best-practice.
1. The Legal Basis of the Online Compliance Intervention Program
1.1 Background on the Program
The operation of Centrelink’s compliance program has been described in detail in the Committee’s
Interim Report. It relied upon the overlapping operation of two Federal Acts: the Data
-
matching
Program (Assistance and Tax) Act 1990 (Cth) (Data-matching Act) and the Social Security Act
1991 (Cth) (Social Security Act). Broadly, the system worked by drawing yearly income
information (primarily from the ATO) through the operation of the Data-matching Act and
comparing this against the income an individual had declared to Centrelink for the period they
had received a relevant social security payment. We note that there is nothing unlawful about
using the data-matching act to gather information to trigger further investigation.
1
Australian Human Rights Commission. ‘Human Rights & Technology Consultation’ (Web Page, 2020)
<https://tech.humanrights.gov.au/consultation>.
4
1.2 Current Legal Questions:
The (Un)lawful Basis of Income Averaging
Following the government’s concession in the case of Amato
2
, it is no longer contested that
issuing debt notices on the basis of ATO income data is unlawful. This point had been made to
government well before the concession in Amato,
3
and we note that the government had, at some
stage, received legal advice regarding the unlawfulness of the scheme. This legal advice has not
been tabled in the Senate despite requests by this Committee. We recommend that the
Committee again call for this information to be revealed, so that further investigation into why the
Program continued to operate unlawfully can be conducted.
The Government has agreed to repay any amounts repaid for the debts unlawfully issued due to
their reliance on ATO income data. The Government has not however agreed to repay interest
on these payments.
Interest Claims
The question regarding whether interest must be repaid is concerned with the overarching
approach Australian courts will take in dealing with claims for restitution due to an ultra vires act
of government. The case of Woolwich if applied by the Court in the present case would likely
require the Commonwealth to provide restitution for the money paid, and any interest arising from
those repayments.
4
Woolwich is a UK authority which was concerned with the question of whether interest needed to
be paid by the Inland Revenue Commission on money collected by a tax that was held separately
by the Court to have been ultra vires. The Court held that restitution was owed with respect to the
money, and that this action accrued as soon as the money was unlawfully collected. As such
interest needed to be repaid (the IRC had already repaid the money unlawfully collected).
We note that the application of the principles discussed in Woolwich would be limited in an
Australian context.
5
This is due to the Australian tax legislation displacing restitution from most
revenue matters.
6
However there does not appear to be a legislative bar in the case of payments
of debts raised purportedly under the Social Security Act. Several commentators have suggested
that there seems to be a prima facie justification for the establishment of a cause of action similar
2
Amato v Commonwealth, VID611/2019. (‘Amato’).
3
See, eg, Terry Carney, ‘The New Digital Future for Welfare: Debts without Legal Proofs or Moral
Authority’ (2018)(1) The New Digital Future for Welfare 1.
4
Woolwich Equitable Building Society v Inland Revenue Commissioners [1993] AC 70 (‘Woolwich’).
5
For discussion see: Janina Boughey, Ellen Rock and Greg Weeks, Government Liability: Principles and
Remedies (LexisNexus, 2019) ch 16. See especially [16.4]-[16.5].
6
Ibid 502 citing Commissioner of State Revenue (Vic) v ACN 005 057 349 Pty Ltd (2017) 261 CLR 509,
535 [73]-[74] (Bell and Gordon JJ), 538 [87] (Gageler J).
5
to Woolich, given the fact that decision-making occurs under the authority of legislation.
7
If
restitution is found to be a viable remedy aggrieved persons will, in effect, have a chose in action
against the Commonwealth. This will mean that the Commonwealth will be precluded through the
operation of s 51(xxxi) of the Constitution from extinguishing that right by legislation unless ‘just
terms’ are provided.
8
There are clear public policy arguments in favour of the imposition of a duty to repay both money
received for a decision made ultra vires and any interest accruing on that money. Most notably, it
is unclear why the government should be able to retain any benefit from debt notices issued ultra
vires. We also note that in Amato interest was repaid. It therefore seems incongruous to withhold
interest in relation to the later class action.
We recommend that, regardless of the court action, the Committee call on the government to
agree to pay interest on top of the repayment of any debts issued unlawfully through the use of
ATO data.
Recommendation 2: The Committee should call on Government to agree to repay
interest to individuals who paid money in response to unlawful debt notices.
Debts issued on the basis of voluntary supply of information
While the precise arguments will be played out before the Federal Court, based on the information
contained in the Interim Report, it appears that the Government is contending that the information
requested in the first notices was authorised under s 66A of the Administration Act (‘the Act’). As
such, any debt raised on the basis of information provided is lawful.
This argument, as a matter of policy should be rejected. Where individuals did not reply to the
notices they risked a debt being raised on the basis of averaged income data. As noted above,
this approach is unlawful. Individuals were also likely to face garnishee notices from the ATO or
from Centrelink, and given these significant consequences were effectively compelled to supply
the requested information. The supply of the information should therefore not be regarded as
voluntary.
Further, we mirror the submissions made to the Inquiry by Terry Carney, in that it is unclear how
s 66A authorises the request for information.
9
In particular, as no change in circumstance has
occurred s 66A does not require the supply of additional information. We would make the
additional point that given the presence of specific information gathering powers contained in the
Act, as a matter of statutory interpretation it is unlikely that s 66A would extend to the situations
7
Boughey, Rock and Weeks (n 4) [16.6].
8
Will Bateman, ‘Legislating against Constitutional Invalidity: Constitutional Deeming Legislation’ (2012)
34 Sydney Law Review 721, 748-50.
9
Terry Carney, Submission No 28 to Senate Standing Committee on Community Affairs, Parliament of
Australia, Review of Centrelink’s Compliance Program (27 September 2019).
6
suggested by Centrelink. In particular, if s 66A imposed a positive duty on individuals to correct
the record (so to speak) regarding any alleged debt, then there would be no purpose for the
powers contained in ss 192-198.
Instead, Centrelink should be relying on the operation of ss 192-198 for their information gathering
powers, or amendments should be made to the Act clarifying the extent of the information-
gathering power contained in s 66A.
Recommendation 3: The Committee should consider if, as a matter of policy, the
government should waive all debts arising from the operation of the compliance
program to ensure that future compliance activities are not impacted by a reduced
willingness to cooperate and supply requested information. This should include
debts raised on the basis of information supplied to Centrelink in response to an
initial notice.
Recommendation 4: Any information gathering powers provided by s 66A should
be clarified, and the section should be amended to expressly include the
protections afforded in ss 192-198 (regarding the required standard of the notice).
2. Services Australia's processes and progress in repaying debts based on income
averaging to date
In May 2020, Services Australia made a commitment to repay debts administered under the
Compliance Program raised through ATO data-matching.
10
Individuals who repaid part or all of
the debts raised through averaged income data from the ATO since July 2015 are eligible for
these repayments, with the onus upon Centrelink to repay wrongful debts and no additional
requirement for individuals to contact Centrelink.
11
10
‘Information About Refunds For The Income Compliance Program’, Services Australia (Web Page, 19
August 2020)-<https://www.servicesaustralia.gov.au/individuals/subjects/information-about-refunds-
income-compliance-program#a3>; Luke Henriques-Gomes, ‘Robodebt: government to refund 470,000
unlawful Centrelink debts worth $721m’, The Guardian (online, 29 May 2020)
<https://www.theguardian.com/australia-news/2020/may/29/robodebt-government-to-repay-470000-
unlawful-centrelink-debts-worth-721m>.
11
‘Information About Refunds For The Income Compliance Program’ Services Australia (Web Page, 19
August 2020)-<https://www.servicesaustralia.gov.au/individuals/subjects/information-about-refunds-
income-compliance-program#a3>.
7
However, Services Australia has revealed that a large quantity of these repayments will take
months to be repaid, with some not scheduled to be repaid until November 2020.
12
Additionally,
the department has announced that larger repayments will be paid in installments.
13
We urge the
Committee to consider the current climate. Amidst COVID-19 it is imperative that vulnerable
people have access to their money, particularly those on Centrelink payments and thus, delays
in repaying individuals are unacceptable. In the context of the wider pandemic recovery,
repayments to individuals will offer wider flow-on benefits to the economy given that these
individuals are more likely to spend the repayments than save it.
14
Centrelink revealed that there are more than 470,000 debts that have to be waived and repaid,
that amount to over $700 million.
15
There has been minimal information provided, however, as
to the resources that are being allocated for the repayments or the processes being used. This
limits the accountability of this process, as well as impeding the planning abilities of affected
individuals in organising their finances.
Whilst we strongly support the repayment of wrongfully administered debts under the Compliance
Program, we also submit that Services Australia should commit to repay these debts as efficiently
and as soon as possible to meet the needs of those affected, particularly amidst COVID-19.
Services Australia should also commit to greater transparency in what resources are being
allocated to this repayment process as well as announcing a clear timeline for the process.
Recommendation 5: That Services Australia commit to a more efficient rollout of
debt repayments, and that a greater effort is made to explain what resources and
programs have been committed to this effort.
3. The Impact of the Compliance Program on individuals, the community sector and
Centrelink staff
The negative ramifications of wrongful debts from the Compliance Program have had particularly
grievous impacts on students across Australia. The Department of Social Services in their
submission revealed that 28% of debt reviews concerned Youth Allowance and a further 47%
concerning Newstart, both of which reveal a substantial impact on students and young people in
12
Jordan Hayne and Matthew Doran, ‘Government to pay back $721m as it scraps Robodebt for
Centrelink welfare recipients’, ABC News (online, 29 May 2020)-<https://www.abc.net.au/news/2020-05-
29/federal-government-refund-robodebt-scheme-repay-debts/12299410>.
13
Ibid.
14
Gareth Hutchens, ‘Young and low-income most likely to spend $750 coronavirus stimulus payments:
ANZ economists’, ABC News (online 9 September 2020) <https://www.abc.net.au/news/2020-09-
09/stimulus-cash-payment/12642904>.
15
Henriques-Gomes (n 9).
8
general.
16
The National Union of Students (‘NUS’) in their submission to the present inquiry note
the vulnerability of students with respect to the program considering the prevalence of low-income
students on social support payments, and the variability in student income.
17
The NUS also
highlighted qualitative responses from the #NotMyDebt Campaign which reveal extensive mental
health impacts upon students from being wrongfully administered debts.
18
The NUS submission
discussed how many students indicated ‘perceptions of illegitimacy as it seemed like a scam’.
19
We emphasis the impact of the Compliance Program on students as a student-based research
hub, we note that many other vulnerable community groups have also been affected by the
program. Overall, we propose that Services Australia enunciate a greater consideration of how
social security decisions, especially those concerning debt programs, will affect the wellbeing of
relevant individuals, before changes to these programs are made.
Recommendation 6: That Services Australia should be obligated to consider the
wellbeing impacts on individuals, particularly those most vulnerable, when making
decisions such as for social security payments.
4. The legal basis and best practices when using automated decision-making
4.1 Benefits and Risks of Automated Decision-Making by Government
Automated decision-making offers qualities of efficiency, high precision, cost-effectiveness and
consistency, making it attractive to government, especially as the number (and therefore cost of)
of decisions that need to be made increases.
20
These benefits do however come with costs. The
rapidity of algorithmic decision-making means that simple mistakes in underlying systems can
result in consistently problematic, incorrect or unlawful decisions at unprecedented scales. This
situation occurred with the Centrelink Compliance Program, where an underlying mistake – the
use of averaged income data to generate debt notices – had its consequences magnified through
16
Services Australia (Department of Human Services), Submission No 20 to the Senate Community
Affairs References Committee, Parliament of Australia, Review of Centrelink’s Compliance Program
(September 2019) 21.
17
National Union of Students, Submission No 35 to the Senate Community Affairs References
Committee, Parliament of Australia, Review of Centrelink’s Compliance Program (September 2019) 3.
18
Ibid 6.
19
Ibid.
20
Will Bateman, ‘Algorithmic Decision-Making and Legality: Public Law Dimensions’ (2020) 94 Australian
Law Journal 520.
9
the use of an automated process.
21
Given this risk, the Government needs to take additional steps
to ensure that decisions made by automated systems are lawful, fair and transparent.
The use of automated processes also raises public law issues regarding the generation of
statements of reasons, the accountability of decision-makers and the public accessibility of
decision-making.
22
There are also wider societal risks if individuals do not understand why
decisions are being automated, the reasoning behind these decisions, or view the process as
unfairly harsh. In such circumstances, people can lose faith with government-deployed automated
decision-making, and subsequently government intervention as a whole.
4.2 The Robodebt Debacle and Issues in Accessibility of Information
A confluence of the above factors has led to the debacle that has forced two government inquiries
to-date (with further review already recommended by the Interim Report). We support wider calls
for review concerning the appropriateness of the use of an automated system to generate debt
notices, given the very real human cost faced where debts are incorrectly and unlawfully
generated. We further support recommendations previously made to the AHRC that human
involvement in automated decision-making ‘should be scaled based on the context and potential
impact of the decision’.
23
On the bases of the inadequate quality of the technological systems deployed in this space to
date, and both of the aforementioned recommendations, it seems inapprorpiate for the
government to use largely or wholly automated processes to raise debts now or in the immediate
future. Any resumption of automated debt-raising would likely only be appropriate (and indeed,
able to gain the faith of the Australian populace) if considerable progress in legal regulation of
automated decision-making occurs in Australia, including (but not limited to) consistent and
generalisable frameworks regarding their explainability and interpretability.
With respect to this aim of improving the transparency and reviewability of automated decision-
making (including Artificial Intelligence and Machine Learning), present laws exempting
underlying code from Freedom of Information requests (and indeed Senate review) are
concerning.
24
Recent research has identified several avenues for advancement in the space of
algorithmic explainability; data-driven techniques (including causality, correlations and example-
based explanations) as well as non data-driven techniques (including the exploration of
implications of a certain model structure and properties of statistical reasoning styles) have been
21
We note that it is still not clear how the underlying system works, and if it deploys any form of Artificial
Intelligence or Machine Learning. However, in absence of evidence to the contrary, simple automation
algorithms combined with the problematic income averaging approach can account for the behaviour of
the program, without an obvious need for these other techniques’ involvement.
22
See, eg, Bateman (n 19); See also Andrew Ray, ‘Implications of the future use of machine-learning in
complex government decision-making in Australia’ (2020) 1(1) ANU Journal of Law and Technology 4.
23
See, eg, Allens Hub for Technology Law and Innovation, ‘Submission to the AHRC Discussion Paper
on Human Rights and Technology’ (17 March 2020).
24
See, eg, Ashlynne McGhee, ‘Centrelink debt recovery program: Department rejects FOI requests
relating to plagued scheme’, ABC News (online, 10 February 2017) <http://www.abc.net.au/news/2017-
02-10/centrelink-debt-recovery-programfoi-requests-rejected/8258564>. For discussion see Ray (n 21)
13-14.
10
argued to play important, distinctive and indeed complementary roles in a general framework of
explainability.
25
However, a typical requirement of these approaches is access to either the
underlying model, the data used in model training, validation, and/or inference, or both. Future
reviews of Freedom of Information laws in this space, therefore, must be alive to these
requirements if generally applicable frameworks of explainability are to be adopted.
26
Additional legal tensions exist regarding the public’s access to information in this space. Namely,
refusing individuals access to the underlying decision-making system potentially undermines
existing legal rules that prohibit policy from altering legislatively applied rules.
27
Further, the
Freedom of Information Act prohibits executive decision-makers from relying on policy to the
detriment of an individual where that policy is not published.
28
Automated decision-making,
particularly where it uses some form of machine-learning algorithm, may be captured by these
provisions.
4.3 A broad review?
We recommend, because of the failings of algorithmic decision-making seen within the Centrelink
Compliance Program, a significant review be undertaken, that address the need for concrete
regulations that can hold the government accountable regarding future (and current) deployments
of algorithmic decision-making systems. Two key elements of such a review should include
enforceable standards of explainability and interpretability in methods deployed by the
government, both in the context of debt repayment and more generally; as well as alternative
preemptive regulatory approaches.
To the best of our knowledge the Robodebt algorithm was not overly complex, relative to the
state-of-the-art deep learning algorithms that are the subject of broader academic debate
regarding the use of AI decision-making. Despite the relative transparency of the structural and
normative problems at play in this algorithm’s behaviour (the misguided reliance on ‘income
averaging’), in the time that it took for its problems to be properly realised and its use halted, it
had already adversely impacted hundreds of thousands of Australians. It is our opinion that this
fact holds deeply troubling implications for the government’s potential to successfully deploy other
more opaque algorithmic decision-making processes going forward. This risk is heightened if
Intellectual Property laws and exemptions to Freedom of Information requests continue to restrict
public scrutiny.
However, as acknowledged above, automated processes have the potential to alleviate burdens
on government, therefore the government’s focus should be the support of research and
development in the space of Artificial Intelligence and Machine Learning
explainability/interpretability and algorithmic regulation to overcome these dangers. This is
25
Atoosa Kasirzadeh, ‘Mathematical decisions and non-causal elements of explainable AI’ (2019)
Workshop on Human-centric Machine Learning, NeurIPS, 24.
26
As we later discuss, this should be viewed as a prerequisite for the use of automated decision-making
by government.
27
See Ray (n 21).
28
Freedom of Information Act 1982 (Cth) ss 8-10.
11
instead of the abandonment of automated decision-making altogether. We acknowledge that this
space is inherently a broad one, but this does not justify an ad hoc approach to the regulation of
algorithmic decision-making. Instead, it suggests that a generalised framework will necessarily
require a (potentially complex and sophisticated) amalgamation of diverse approaches to
regulation. The complexity of the required regulation should not prevent the creation of a
consistent and enforceable process of scrutiny.
We nominate numerous broad avenues of legal and technical research and development to which
the government could explore to potentially incorporate into regulatory legislation in this space.
Firstly, we encourage broad and holistic approaches to the issue of mandating algorithmic
explainability and interpretability. This includes the acknowledgement that many approaches to
explanations are required for sufficient interpretability. This includes data-driven and non-data-
driven techniques, which can range from explaining salient factors in a specific algorithmic
decision to explaining the underlying assumptions of a statistical and optimisation-based
approach.
29
The continued development of techniques for effectively communicating potentially
complex ideas to the public in all of these approaches is crucial, as is ensuring new techniques
are included in concrete regulation as they emerge.
We particularly recommend the support of further research into ‘model-agnostic’ explanatory
techniques.
30
These facilitate interpretable explanations of complex models to the public.
Moreover, such techniques also provide a potential remedy for the tension between regarding
Freedom of Information requests and/or Intellectual Property claims and government efficiency,
by allowing sufficient explanatory information about the model’s behaviour to be released, whilst
the actual underlying implementation remains unavailable.
Finally, we encourage investigation of regulatory options beyond explainability/interpretability, as
well. Regulatory approaches such as ‘Data Protection Impact Assessments’ and certification
schemes, which can make a concrete contribution to more responsible deployment of automated
decision making, have emerged overseas.
31
Our government has the opportunity to observe their
potential successes and adopt similar approaches.
29
Kasirzadeh (n 24) 24.
30
Scott M. Lundberg and Su-In Lee, ‘A Unified Approach to Interpreting Model Predictions’ (2017) 31st
Conference on Neural Information Processing Systems; Marco Tulio Ribeiro, Sameer Singh and Carlos
Guestrin, ‘“Why Should I Trust You?” Explaining the Predictions of Any Classifier’ (2016) Knowledge
Discovery and Data Mining (ACM KDD).
31
Lilian Edwards and Michael Veale, ‘Enslaving the Algorithm: From a “Right to an Explanation” to a
“Right to Better Decisions”?’ (2018) 16(3) IEEE Security & Privacy, 46-54.
12
Recommendation 7: A review of algorithmic decision-making’s appropriateness in
the specific context of debt-raising should precede its resumption in any form.
Recommendation 8: Humans should maintain involvement in decision-making
where those decisions will have a significant impact on individuals.
Recommendation 9: The Freedom of Information Act should be amended to make it
easier for individuals (and the Senate) to access information relating to the use of
automated systems by government.
Recommendation 10: The human costs from the Compliance Program’s use should
motivate a strengthened commitment to government support and consultation of
both technical research into algorithmic explainability/interpretability, and legal
research into effective algorithmic regulation.
Recommendation 11: This research should be prioritised commensurate to the
government’s intent to deploy and rely on automated decision-making systems.
Recommendation 12: The government should prioritise the creation of generally
applicable standards of automated decision-making, taking in the rapid advances
of AI and machine-learning technology.
5. The future of Centrelink compliance activities and programs, particularly
compliance activities for the recovery of social security overpayments
Centrelink appropriately paused compliance activities due to the ongoing impact of COVID-19.
We agree with comments made by the Interim Report that this pause should be regarded as the
perfect time for Centrelink to review its activities and assess the lawfulness of its decision-making
practices.
Given our discussion above, we further recommend that regardless of the lawfulness of the
current compliance program
32
the Department should not resume its compliance program in its
current form. This is supported by the ongoing impact regarding the use of automated decision-
making and shifting of burden of proof on individuals. While the recovery of public money that has
been mistakenly or fraudulently received is an important objective of government, so to is
protecting vulnerable individuals affected by what could best be described as a flawed and
overeager compliance system. As has been demonstrated, rushing to recover these “debts” does
not necessarily benefit the Consolidated Revenue Fund, with significant resources being diverted
to fighting the ongoing class action as well as by inquiries to review the compliance program.
32
For example, regardless of whether the present notices are being lawfully issued under s 66A.
13
These costs will be further exacerbated if the Committee is successful in its calls for a Royal
Commission into the practice.
As already highlighted above, it may be necessary to amend social security legislation to clarify
the extent of powers currently authorised under statute to enable an open and transparent review
of the compliance activities. Instead we recommend that future compliance activities are based
on the below principles:
1. That the benefit in preventing fraud is balanced against the harm of the relevant
compliance program;
2. Income averaging is only used to identify social security recipients for further investigation
by a human decision-maker;
3. That the decision to raise a debt against an individual is made in a lawful, transparent and
open manner;
4. In particular, where an automated process has been used at any stage of the process the
individual is notified of that process;
5. Care needs to be taken to ensure that individuals are kept up-to-date about what is
occurring, in this context the use of old address information held by Services Australia is
of concern. Where possible, notification should be made through MyGov in addition to the
latest postal address – which will likely be held by the ATO.
6. Where information is requested from an individual, that the request is lawful and complies
with the relevant statutory obligations;
7. Where a notice is issued, individuals are told the specific legislative basis authorising the
notice;
8. Where a decision is made, individuals are told in clear and simple language their options
for internal and external review of the decision.
Recommendation 13: Future compliance activities adopt a principles-based
approach to debt recovery.
Recommendation 14: Any further review of the compliance program should
consider the complexity in the Acts underpinning its operation and amend those
Acts to improve the openness and transparency of decision-making practices.
We would be happy to answer questions, or provide further submissions if requested by the
committee and may be contacted at anulrsjresearchhub@gmail.com.
Yours sincerely,
Andrew Ray, Annika Reynolds, Charles Evans, Pippi Callan, Sathyajith Sukumaran, Sebastian
Judge