Human Technology Institute submission on ADM systems (Provider advocacy)

‹ PrevPage 1 of 19 · Source p. 1Next ›

Submission 347

29 May 2026

Human Technology

Institute

HumanHuman TechnologyTechnology

InstituteInstitute

TEXTTEXT

Inquiry into the National Disability Insurance Scheme

Amendment (Securing the NDIS for Future Generations) Bill 2026 Inquiry into the

Submission to the Senate Community Affairs Legislation

Committee National Disability

29 May 2026 I S h

1

Submission 347

29 May 2026

The Human Technology Institute (HTI) is building a future that applies human values to new technology. HTI embodies the strategic vision of the University of Technology Sydney (UTS) to be a leading public university of technology, recognised for its global impact specifically in the responsible development, use and regulation of technology. HTI is an authoritative voice in Australia and internationally on human-centred technology. HTI works with communities and organisations to develop skills, tools and policy that ensure new and emerging technologies are safe, fair and inclusive and do not replicate and entrench existing inequalities.

The work of HTI is informed by a multi-disciplinary approach with expertise in data science, law and governance, policy and human rights.

For more information, contact us at hti@uts.edu.au

Authors: Clare Brennan, Sarah Sacher and Prof Edward Santow

Acknowledgement of Country

UTS acknowledges the Gadigal people of the Eora Nation, the Boorooberongal people of the Dharug Nation, the Bidiagal people and the Gamaygal people upon whose ancestral lands our university stands. We would also like to pay respect to the Elders both past and present, acknowledging them as the traditional custodians of knowledge for these lands.

2

Submission 347

29 May 2026

Executive summary 1 List of Recommendations 3 Overarching comments and recommendations 5 Need for legislated framework regulating ADM use in government 5 Delegation to computer provisions should be made consistent 5 Specific comments and recommendations on Schedule 3, Part 2, Division 5 of the Bill 6 Automation of ‘evaluative administrative actions’ 6 Automation of administrative action by legislative instrument 10 CEO substitution power 11 Hidden limitations on internal and external review 12 Need for additional safeguards 13 Implications for NDIS culture and staff 15

3

Submission 347

29 May 2026

Executive summary

The Human Technology Institute (HTI) welcomes the opportunity to submit to the

Senate Community Affairs Legislation Committee’s inquiry into the National Disability

Insurance Scheme Amendment (Securing the NDIS for Future Generations) Bill 2026 (the Bill). The Bill would amend the National Disability Insurance Scheme Act 2013 (Cth) (the Act).

HTI’s submission focusses on Schedule 3, Part 2, Division 5 of the Bill (the ‘automation provisions’) which would enable the CEO to automate a broad range of administrative action, including many types of administrative decision.

Thoughtful adoption of automated decision-making (ADM) systems within the NDIS has a range of potential benefits. The Explanatory Memorandum states that the automation provisions in the Bill are intended to assist the NDIS to provide ‘high quality, efficient and timely services.’1 These are important goals. The use of ADM in the NDIS could support the achievement of these goals, but only if three vital conditions are satisfied:

  • the legal and governance framework underpinning the use of ADM must be robust

  • any ADM system that is used in the NDIS should be designed and deployed to ensure that the combination of the technology and the humans responsible for overseeing the technology delivers high-quality, reliable decisions

  • the accountability protections, which enable individuals to appeal erroneous or otherwise problematic decisions, must be straightforward, accessible and effective at addressing such problematic decision making.

A failure to meet one of these conditions is likely to compromise the relevant decision making system. As explained in greater detail in the body of this submission, HTI holds grave concerns that the automation provisions of the Bill fail to satisfy all three of these conditions.

HTI’s central concern with the Bill is as follows. The automation provisions in the Bill are facilitative: they provide legal authorisation to automate some forms of NDIS decision making, with a breadth that is, to our knowledge, unprecedented in Australian law. Such legislation cannot be a blank cheque. In authorising some forms of ADM, it must also make clear what is and is not permissible in an area of decision making that engages fundamental human rights.

Like human-made decisions, decisions made or materially influenced by ADM must be procedurally fair, accurate, transparent and subject to genuine oversight, with associated rights of review.2 A person affected by an automated decision should understand why the decision was made, and how to challenge a decision. Decisions and actions must not arbitrarily interfere with a person’s human rights, and they must be non-discriminatory. Sometimes the use of ADM can make those basic legal duties more difficult to comply with, or ADM can result in decisions whose lawfulness or otherwise is difficult or impossible to assess. In those scenarios, ADM may simply be inappropriate.3

1 Explanatory Memorandum, National Disability Insurance Scheme Amendment (Securing the NDIS for Future

Generations) Bill 2026 (Cth) 126. 2 See Commonwealth Ombudsman, Automated decision-making: Better practice guide (2019) <https://www.ombudsman.gov.au/__data/assets/pdf_file/0029/288236/OMB1188-Automated-Decision-Making Report_Final-A1898885.pdf>. 3 See, eg, Australian Human Rights Commission, Human Rights and Technology (Final Report, March 2021) 82.

1

Submission 347

29 May 2026

By simply facilitating ADM, without setting clear guardrails to protect legal and human rights, this Bill fails in a fundamental way. Decision making in a multifaceted area, like the NDIS, is inherently complex. Errors are inevitable, but the law has an important role in minimising this risk and in enabling redress in the event of error. While the Bill provides some minor safeguards associated with notice and transparency, it simultaneously rolls back more fundamental protections for human discretion, oversight, responsibility for decision making, and meaningful access to review.

The NDIS exists to support people with disability, including those experiencing significant vulnerability, marginalisation and disadvantage. It is well understood that people in these groups are more likely to be disproportionately harmed or poorly served by ADM systems, especially where they rely on government systems to meet basic needs.4 People with disability require human support to navigate complex systems and have their needs understood. If this impact is not adequately considered when ADM is adopted, the downstream harms can be catastrophic, and in some cases deadly – as illustrated by the experience of Robodebt.

In considering a major expansion of ADM in the NDIS, Parliament must learn from the experience of other ADM initiatives introduced by the Australian Government. For example:

  • The integrated assessment tool (IAT), introduced by the Department of Health, Disability and Ageing in July 2024, is used to assess funding packages for older Australians accessing aged care services. Human assessors are required to accept the system generated outcomes that that the IAT provides.5 As of March 2026, there had been 834 review requests,6 accompanied by a 50% surge in demand for advocacy services since its introduction.7 The IAT is currently being investigated by the Commonwealth Ombudsman.8

  • The Targeted Compliance Framework, administered by the Department of Employment and Workplace Relations (DEWR), has been plagued with problems associated with inaccurate suspension and cancellation of jobseeker payments through automation. In 2025, the Commonwealth Ombudsman found that DEWR’s cancellation of income support was not lawful.9 The Framework has been flagged as an ‘unquantifiable liability’ in this year’s budget.10

  • In 2024, the Australian Tax Office (ATO) sent up to 200,000 boilerplate, automated letters to Australian tax payers and tax agents, stating that they had ‘on-hold’ historical tax debts. The letters did not provide detail about how or why the debts had arisen, and caused confusion and distress to recipients. The ATO publicly apologised.11

4 See, eg, Virginia Eubanks, Automating Inequality: How high-tech tools profile, police and punish the poor (St Martin’s Press, 2018); Philip Alston, ‘Report of the Special Rapporteur on extreme poverty and human rights’, UN Doc A/74/493 (1 October 2019). See generally, Royal Commission into the Robodebt Scheme, Final Report (July 2023). 5 Parliament of Australia, Parliamentary Debates, Senate, 11 February 2026, 19 (Ms Rachel Blackwood). 6 Evidence to the Community Affairs References Committee, Parliament of Australia, Canberra, 1 April 2026, 6 (Robert Day). 7 Melissa Davey, ‘“Stuff of nightmares”: calls for help surge 50% after Australia launches aged-assistance tool’ The Guardian, 24 February 2026 <https://www.theguardian.com/australia-news/2026/feb/23/calls-for-help-surge-aged assistance-tool-care-homes-australia>. 8 Melissa Davey, ‘Labor’s controversial algorithm tool for aged care under investigation by Ombudsman’ The Guardian, 15 April 2026 <https://www.theguardian.com/australia-news/2026/apr/14/labor-aged-care-algorithm-tool-ombudsman investigation>. 9 Commonwealth Ombudsman, Automation in the Targeted Compliance Framework: when the law is changed but the system isn’t (August 2025) 5 <https://www.ombudsman.gov.au/__data/assets/pdf_file/0017/320750/Automation-in-the Targeted-Compliance-Framework.pdf>. 10 Joseph Brookes, ‘Automated welfare compliance saga hangs over budget’ Innovation Aus https://www.innovationaus.com/automated-welfare-compliance-saga-hangs-over-budget/. 11 Jonathan Barrett, ‘’It doesn’t make sense’: why recipients of ATO ‘robotax’ letters are scratching their heads’ The Guardian (30 November 2023) <https://www.theguardian.com/australia-news/2023/nov/30/ato-on-hold-tax-debt-letters

2

Submission 347

29 May 2026

  • The Australian National Audit Office’s review into Services Australia’s ISIS system, used to calculate fortnightly pension rates, found that between July 2021 and June 2024 there was an estimated total underpayment of $1.33 billion, and $3.67 billion in overpayments.12

List of Recommendations

Recommendation 1

The Australian Government should urgently implement a legislated framework for use of automated decision making in government, drawing on HTI’s proposed model in its 2025 submission to the Attorney-General’s Department.

Recommendation 2

(a) Before introducing a computer-delegation provision to Parliament, the Australian Government should clearly articulate why the delegation is appropriate under the relevant Act. Any computer-delegation provision in the NDIS Act should be drafted to make clear:

o precisely which categories of decision and administrative action the provision applies to

o which officers of the Commonwealth (as this term is understood under the Constitution) are legally responsible for the decision or action

o what safeguards apply to ensure that decisions are made fairly and lawfully.

(b) The Australian Government should conduct an audit of existing computer delegation provisions, assessing these provisions against the criteria in the

Commonwealth Ombudsman’s Better Practice Guide on Automated Decision

Making. Amendments should be made where required.

Recommendation 3

The Bill should be amended to provide that, in taking administrative action, the exercise of discretion, or the formation of an evaluative judgment or a particular state of mind, generally should not be automated. This principle need not apply where the exercise of discretion is for the clear and unambiguous benefit of the affected person.

Recommendation 4

(a) The Bill should be amended to remove proposed s 59C(2) of the NDIS Act, which provides that the Minister may ‘specify’ provisions for automation through legislative instrument.

(b) If Parliament does not make the amendment proposed in Recommendation 4(a) above, the Bill should be amended to set guardrails for how the Minister can exercise the power in proposed s 59C(2). Those guardrails should include: that the Minister may not specify provisions that involve the exercise of discretion or evaluative judgment; and that the Minister must be reasonably satisfied that the legislative instrument does not in substance, reduce an affected individual’s rights to internal or external review.

robodebt-apology>; Jonathan Barett, ‘ATO apologises for ‘unnecessary distress’ caused by on-hold tax debt recovery letters’ (29 November 2023) <https://www.theguardian.com/australia-news/2023/nov/29/ato-apologises-for unnecessary-distress-caused-by-on-hold-tax-debt-recovery-letters>. 12 Australian National Audit Office, ‘Administration of the Age Pension’ (Auditor-General Report No 20 of 2025-26) https://www.anao.gov.au/work/performance-audit/administration-of-the-age-pension.

3

Submission 347

29 May 2026

(c) Where administrative action is to be automated, this should be provided in primarily legislation, not through delegated legislation.

Recommendation 5

(a) The Bill should be amended to include a provision enabling decisions to be substituted by officers for a correct or preferable decision, in circumstances where the substitution will not compromise the integrity of the initial decision (i.e. will not be less accurate, or contrary to human rights principles).

(b) Sched 3, Part 2, Item 12 (proposed s 202(2A) of the NDIS Act) should be deleted.

Recommendation 6

(a) Proposed s 59B(11) of the Bill should be amended to:

o preserve rights of internal and external merits review for decisions made under designated or specified provisions; and

o provide that any administrative decision made or materially influenced by ADM under the Act should be subject to external merits review in circumstances where the decision would be reviewable if a human decision maker had made it.

(b) The NDIA should also set up accessible pathways for seeking internal review of automated decisions made under the Act.

Recommendation 7

(a) The Bill should be amended to include additional safeguards for all automation of administrative actions under the Act, where automation would have the potential to reduce individuals’ legal or human rights.

(b) These safeguards should include specified requirements for: genuine human oversight; reasons for decisions; notification of ADM errors; testing of ADM systems; public information about the technical workings of ADM systems; external oversight of ADM systems; and consideration of human rights in the development of administrative instruments.

Recommendation 8

(a) The NDIA should invest in training NDIS staff so they can effectively review ADM decisions, with an understanding of ADM technology they work with, the underlying law, and the application of human-centred AI and ADM principles.

(b) The NDIA should consult with NDIS staff before adopting ADM systems that staff will interact with, or rely upon, to make decisions.

4

Submission 347

29 May 2026

Overarching comments and recommendations

Need for legislated framework regulating ADM use in government

The Robodebt Royal Commission recommended the Government consider legislative reform to introduce a consistent legal framework in which automation in government services can operate.13 A consultation process was conducted by the Attorney General’s Department in December 2024 on an ADM framework for government, but since then there has been no report or recommendations released, or any public communications made on the subject.

HTI refers the Committee to its submission to this consultation process, which sets out a proposed model for a legislated ADM framework, and reiterates its call for an ADM framework to be implemented as an urgent priority.14

An ADM legal framework would provide for the consistent application of administrative law and human rights principles to ADM use across government, with a standard set of safeguards and accountability mechanisms. This would: address problems with agencies each independently developing their own rules; provide much needed certainty about the application of the law to ADM; and embed a preventative approach to avert harms before they arise. For example, an ADM legal framework would ensure that agencies apply risk assessments and consider harms and lawfulness from the outset, instead of moving to adopt problematic ADM systems, and seeking statutory authorisation after they are already in train.

Where government introduces ADM without clear legal safeguards, as is the case with this Bill, the law cannot perform its conventional role in mitigating risks associated with ADM. This means that those risks are likely to remain unaddressed, especially at the system level, until they become very large. By that point, the economic and human cost in addressing these risks is far greater than if a more prudent, prophylactic approach had been taken at the outset.

Recommendation 1

The Australian Government should urgently implement a legislated framework for use of automated decision making in government drawing on HTI’s proposed model in its 2025 submission to the Attorney-General’s Department.

Delegation to computer provisions should be made consistent

The Bill includes a set of provisions that are sometimes known as ‘computer delegation’ provisions. As of 2024, 46 separate provisions of this kind have been included in primary legislation, to enable administrative decisions to be made by ‘a computer’ for the purpose of that Act.15 These provisions have been introduced ad hoc, without consistent or sufficient (and in many cases, any) safeguards. This was raised as a concern in the Robodebt Royal Commission report.16

HTI has previously recommended that these provisions be audited and updated to ensure a consistent approach, with appropriate safeguards. When new provisions are introduced, the Australian Government should articulate why the delegation is

13 Royal Commission into the Robodebt Scheme, Final Report (July 2023) Recommendation 17.1. 14 Human Technology Institute, Submission to the Attorney-General’s Department consultation on Use of Automated Decision-Making by Government (January 2025) <https://utsd8.prod.acquia-sites.com/sites/default/files/2025-01/HTI submission-Use-of-ADM-by-Government.pdf>. 15 Attorney-General’s Department, Use of automated decision-making by government (Consultation Paper, November

  1. 5 < https://consultations.ag.gov.au/integrity/adm/user_uploads/consultation-paper-use-of-automated-decision- making-by-government.pdf>. 16 Royal Commission into the Robodebt Scheme, Final Report (July 2023) 485.

5

Submission 347

29 May 2026

appropriate, and they should be drafted by reference to a standard set of criteria and safeguards. HTI suggests adapting the criteria in the checklist included in the

Commonwealth Ombudsman’s Better Practice Guide on Automated Decision Making

for these purposes17

Recommendation 2

(c) Before introducing a computer-delegation provision to Parliament, the Australian Government should clearly articulate why the delegation is appropriate under the relevant Act. Any computer-delegation provision in the NDIS Act should be drafted to make clear:

o precisely which categories of decision and administrative action the provision applies to

o which officers of the Commonwealth (as this term is understood under the Constitution) are legally responsible for the decision or action

o what safeguards apply to ensure that decisions are made fairly and lawfully.

(d) The Australian Government should conduct an audit of existing computer-delegation provisions, assessing these provisions against the criteria in the Commonwealth Ombudsman’s Better Practice Guide on Automated Decision Making. Amendments should be made where required.

Specific comments and recommendations on Schedule 3, Part 2, Division 5 of the Bill

Automation of ‘evaluative administrative actions’

Importance of discretion and evaluative judgment for fair decision making

The most radical element of the Bill’s automation provisions is that they explicitly enable automation of discretionary decisions. This is radical because it would expressly authorise ADM for decision making that goes well beyond routine or purely fact-based decision making.

Explicit legislative authorisation for automated systems to apply to discretionary decisions is, to our knowledge, unprecedented in Australian law. Proposed s 59B(4) enables the CEO to arrange the use of computer programs that involve a discretion being exercised, an evaluative judgment being made, or a state of mind being formed.

To illustrate the distinction between discretionary and non-discretionary decision making, imagine the following hypothetical. A law provides that, if an individual is listed as an Australian citizen on a particular database, they are entitled to a specific government benefit. In this scenario, there is no real discretion in determining whether an individual is entitled to the benefit; the sole criterion is whether they are listed on the relevant database. A human could check the database manually for the individual’s name, or this could be done automatically. Either way, entitlement to the benefit turns solely on whether the individual’s name is listed on the database. Because automatically checking databases can generally be done more efficiently and at least

17 See Commonwealth Ombudsman, Automated decision-making: Better practice guide (2019) 49-58 <https://www.ombudsman.gov.au/__data/assets/pdf_file/0029/288236/OMB1188-Automated-Decision-Making Report_Final-A1898885.pdf>.

6

Submission 347

29 May 2026

as accurately as a manual process, this is the sort of decision that, typically, is most conducive to ADM.

By contrast, now imagine that the law provides that Australian citizens, who would otherwise be at risk of poverty, are eligible for the relevant benefit. In this scenario, there are two criteria: the first is non-discretionary (is the individual an Australian citizen?), and the second is discretionary because it requires an evaluative judgment to be made about the likelihood that the individual could become impoverished. It has long been recognised that the discretionary or evaluative character of the second criterion is more problematic in ADM. This is because relatively simple ADM systems, which rely on algorithms involving the rigid application of pre-defined categories and procedures, flatten individual characteristics and cannot take into account unusual characteristics or outliers. There are problems also with ADM systems that use more sophisticated forms of AI, as those systems rely on patterns in data to make decisions, and are rarely capable of offering a nuanced assessment of individual circumstances – no matter how sophisticated the system.

Returning to the Bill, proposed s 59(C)(1) sets out a list of ‘designated’ provisions to which this power will apply. In addition to the ‘designated’ provisions in the Bill, proposed s 59C(2) would provide that the Minister may, by legislative instrument, ‘specify’ any provision under the Act or a legislative instrument made under the Act to be delegated to a computer (inclusive of evaluative administrative actions).

The automation of the exercise of discretion and evaluative judgment can threaten fundamental principles of administrative law, and the rule of law more broadly.

The exercise of discretion is a crucial component of fair decision making, as the Commonwealth Ombudsman has explained:

[Discretions] are a tool to avoid unfair or unjust outcomes that might otherwise result from inflexible application of a particular statutory provision and can help ensure decision-making is consistent with broader legislative and policy intent.18

Discretionary and evaluative decisions necessarily involve contextual interpretations and nuanced, individualised assessments. Generally speaking, a human decision maker is better placed to make such assessments, because they can assess a particular case on its own merits, in light of the specific factual circumstances that arise, and the individual in question. They are therefore fundamentally ill-suited to automation. Even where human decision makers are assisted by ADM tools to make their own decisions, they may over-rely on system outputs, or have little practical room to manoeuvre due to operational or system constraints.

For these reasons, authorities including the Commonwealth Ombudsman and the Australian Human Rights Commission have urged government to be cautious in automating discretionary decisions.19 The NSW Ombudsman, in its New Machinery of Government Report, underlined that:

If the introduction of [ADM] into a discretionary decision-making system has the effect that the administrator is no longer able to – or does not in practice –

18 Commonwealth Ombudsman, Automated decision-making: Better practice guide (2019) 10 <https://www.ombudsman.gov.au/__data/assets/pdf_file/0029/288236/OMB1188-Automated-Decision-Making Report_Final-A1898885.pdf>. 19 Commonwealth Ombudsman, Automated decision-making: Better practice guide (2019) 10 <https://www.ombudsman.gov.au/__data/assets/pdf_file/0029/288236/OMB1188-Automated-Decision-Making

Report_Final-A1898885.pdf> ; Australian Human Rights Commission, Human Rights and Technology (Final Report,

March 2021) 56.

7

Submission 347

29 May 2026

continue to exercise genuine discretion, that system will be inconsistent with the statute that granted the discretion, and its outputs will be unlawful.20

Additionally, both the Parliamentary Joint Committee on Human Rights and the Senate Scrutiny of Bill Committee have raised concerns regarding computer delegation provisions that apply to administrative decisions involving discretionary elements.21 The Office of Parliamentary Counsel, in its Instructor Guidance Note for agencies on automated decision making, similarly emphasises the risks associated with evaluative decisions and recommends that agencies seek advice from the Attorney-General’s Department or legal advice before developing legislation that may impact upon them.22

Implications of these provisions for the NDIS and NDIS participants

The designated provisions in the Bill, as well as those flagged for potential specification by the Minister in the Explanatory Memorandum,23 have direct and significant impacts on the legal and human rights of NDIS participants. For example, administrative actions and decisions taken under these provisions determine what a person’s support needs are, and the kind and amount of supports and funding they are entitled to over the course of their plan. Such decision-making processes directly impact NDIS participants’ physical and mental health, their dignity and autonomy, their ability to succeed in education and work, and participate in social and public life. In light of these human rights impacts, it is crucial that decision-making processes take into account individual circumstances.

These considerations are heightened by the vulnerability and diversity of NDIS participants.24 ADM systems used to make decisions about individuals are necessarily reductive – this is particularly the case for people with disability, whose individual experiences, and complex needs, can rarely be represented through decontextualised data or neatly categorised into pre-determined boxes.

While HTI considers that ADM is generally ill suited to the formation of an evaluative judgement, where ADM is nevertheless is used for that purpose, it is necessary for the decision-making process to provide for effective human involvement and oversight. A human decision maker must have the ability to consider individual circumstances and evidence, as well as a range of possible options, and depart from ADM outputs or recommendations when needed. The Explanatory Memorandum states that ‘automating part of an action, such as making a recommendation, with human oversight of the final decision, is an appropriate check and balance to guard against unfettered computer-based decision-making.’ However, there is nothing in the text of the Bill that protects or requires staff-level human oversight and discretion.

Automation of evaluative administrative actions will be subject to a ‘standard operating procedure’ to be developed by the CEO. However, reliance on an administrative instrument is not a sufficient guardrail to protect against the automated exercise of discretion, even if the instrument is restricted to automation of objective elements of a decision-making process, as per proposed s 59D(3). This is especially so given that instrument would be non-disallowable, and therefore not subject to the usual

20 New South Wales Ombudsman, The new machinery of government: using machine technology in administrative decision-making (29 November 2021) 33 <https://cmsassets.ombo.nsw.gov.au/assets/Reports/The-new-machinery-of government-special-report_Front-section.pdf.pdf>. 21 See Joint Committee on Human Rights, Parliament of Australia, Human Rights Scrutiny Report 7 of 2018 (14 August

  1. 12; Joint Committee on Human Rights, Parliament of Australia, Human Rights Scrutiny Report 11 of 2018 (16 October 2018) 78; Senate Standing Committee for the Scrutiny of Bills, Delegated legislation monitor, No 4 of 2024, 5. 22 Office of Parliamentary Counsel, Instructor Note – Automated decision making (November 2025) < https://www.opc.gov.au/sites/default/files/2025-11/Automated%20decision-making.pdf>.

23 Explanatory Memorandum, National Disability Insurance Scheme Amendment (Securing the NDIS for Future

Generations) Bill 2026 (Cth) 133. 24 See e.g. Louisa Well, Morgan Currie & James Stewart, ‘Surveillance, Discretion and Governance in Automated Welfare: the case of the German ALLEGRO system’ Science & Technology Studies (2023) 36(1) 44.

8

Submission 347

29 May 2026

parliamentary scrutiny that applies to delegated legislation. Human discretion and evaluative judgment should be safeguarded in primary legislation, and administrative instruments drafted within those boundaries.

Such safeguards are particularly important in light of the new approach to support needs assessments and the introduction of the I-CAN version 6 tool, which provides insight into how evaluative decisions may be automated by the NDIS in practice. In September 2025, the Australian Government announced a new model for conducting support needs assessments to be introduced in mid-2026, to simplify processes.25 Assessments would be based on I-CAN version 6. Like the previous model, assessments would be made through staff interviews with the individual, assisted by the I-CAN tool. Unlike the previous model, staff would no longer have discretion to amend funding and support plans generated by I-CAN. An assessor would merely be able to request that a new assessment be made using different inputs.26

If the new assessment model is implemented in the manner described above, it would significantly reduce the extent to which an NDIS participant’s individual needs can be assessed, and risks disastrous consequences for some NDIS participants, and the integrity of the NDIS itself. Reports that the tool has not been tested in respect of people with a variety of disability types compounds this risk.27

When the I-CAN v 6 tool was first reported on, an NDIA spokesperson stated it is ‘not an automated system. Assessments will be conducted by trained and accredited assessors’.28 The I-CAN tool is the very definition of an ADM system – the involvement of a human reduces automation only if the human has real power over the outcome.

Proposed amendment

The Bill should provide that categories of administrative action involving a discretion being exercised, an evaluative judgment being made, or a state of mind being formed should not be subject to automation as a general rule. If objective parts of a decision making process leading up to the exercise of discretion are to be automated, those elements should be spelled out in primary legislation – and subject to full parliamentary scrutiny.

A narrow exception may be made for identified circumstances where an automated exercise of discretion is for the clear and unambiguous benefit of the affected person. Notably, every example given in the Explanatory Memorandum for the automation of administrative actions involves the rejection of claims – rejections must be subject to more human oversight, not less, because of the negative impact on people.

Recommendation 3

The Bill should be amended to provide that, in taking administrative action, the exercise of discretion, or the formation of an evaluative judgment or a particular state of mind, generally should not be automated. This principle need not apply

25 Sarah Basford Canales, ‘New NDIS needs assessments will use technology to simplify processes but advocacy groups cautious about change’ The Guardian, 25 September 2026 <https://www.theguardian.com/australia news/2025/sep/25/new-ndis-assessments-using-technology-simplify-process-advocacy-groups-hesitant-about-change>. 26 Kate Lyons, ‘NDIS plans will be computer-generated, with human involvement dramatically cut under sweeping overhaul’ The Guardian, 3 December 2025 < https://www.theguardian.com/australia-news/2025/dec/03/ndis-plans computer-generated>. 27 Kate Lyons ‘NDIS tool to determine support not tested on variety of disability types – including diverse autism, experts warn’ The Guardian, 21 January 2026 < https://www.theguardian.com/australia-news/2026/jan/20/ndis-tool-to determine-support-not-tested-on-variety-of-disability-types-including-diverse-autism-experts-warn>. 28 Kate Lyons, ‘NDIS plans will be computer-generated, with human involvement dramatically cut under sweeping overhaul’ The Guardian, 3 December 2025 < https://www.theguardian.com/australia-news/2025/dec/03/ndis-plans computer-generated>.

9

Submission 347

29 May 2026

where the exercise of discretion is for the clear and unambiguous benefit of the affected person.

Automation of administrative action by legislative instrument

Proposed s 59C(2) would provide that a Minister may, by legislative instrument, ‘specify’ any provision under the Act or a legislative instrument made under the Act to be delegated to a computer. This means that any and all administrative actions and decisions made by the NDIS could be automated in the future, and in theory, enables total automation of the NDIS. This power to specify types of decision for automation is very broad: proposed s 59C(3) specifies only that if a provision involves evaluative determinations, the Minister must be satisfied that it is ‘appropriate’ to specify it.

Enabling certain administrative decisions in the NDIS to be automated carries enormous risk, especially given that such decisions engage fundamental human rights. This goes to the very heart of the NDIS. This specification merits the close attention of Parliament – hence, it should be addressed directly in primary legislation. This should not be a matter that is delegated to the Minister, with the consequent reduction in democratic scrutiny.

The Explanatory Memorandum states that the reason why certain provisions haven’t been designated in the text of the Bill is because it is not clear how automation will proceed: ‘for example, the role of a computer system to take or assist a human in an administrative action may be uncertain because the design and build of the computer system may be underway.’29 This abrogates Parliament’s supervisory responsibility on a fundamental question: the appropriateness of the use of ADM in a particular decision making context under the NDIS Act turns, in no small part, on the demonstrated ability of the proposed ADM system to perform appropriately. If the Government cannot demonstrate this, it is inappropriate to designate categories of decisions as being suitable for automation.

If a power for the Minister to specify provisions through legislative instrument is retained in the Bill, the Bill should at least set clear boundaries for the kinds of administrative action that can be specified. For example, the Minister should be restricted from enabling the automation of administrative action involving discretion or evaluative judgment, and there should be a requirement for the Minister to be reasonably satisfied that the legislative instrument does not, in substance, reduce an affected individual’s rights to internal or external review.

Recommendation 4

(d) The Bill should be amended to remove proposed s 59C(2) of the NDIS Act, which provides that the Minister may ‘specify’ provisions for automation through legislative instrument.

(e) If Parliament does not make the amendment proposed in Recommendation 4(a) above, the Bill should be amended to set guardrails for how the Minister can exercise the power in proposed s 59C(2). Those guardrails should include: that the Minister may not specify provisions that involve the exercise of discretion or evaluative judgment; and that the Minister must be reasonably satisfied that the legislative instrument does

29 Explanatory Memorandum, National Disability Insurance Scheme Amendment (Securing the NDIS for Future

Generations) Bill 2026 (Cth) 133.

10

Submission 347

29 May 2026

not in substance, reduce an affected individual’s rights to internal or external review.

(f) Where administrative action is to be automated, this should be provided in primarily legislation, not through delegated legislation.

CEO substitution power

Proposed s 59B(7) of the Bill provides that the CEO may substitute administrative actions when satisfied that the administrative action taken by the computer program is not correct or preferable. This is a non-delegable power per proposed s 202(2A) of the NDIS Act.

Because this power is non-delegable, the apparent effect would be that NDIS staff will not be able to independently correct inaccurate or unfair decisions made by computer systems under the Bill, or exercise relevant discretions in the first instance.

Discretion should be exercised by human decision makers when a decision is first made. Individuals should not have to seek review of a decision in order for discretionary considerations to be applied to them. Not only would this be contrary to administrative law principles, but any harm caused by the original decision will have already been suffered by the individual, and most will not have the knowledge or resources to seek review when discretionary considerations have not been applied.

There is also no requirement in the Bill for the CEO to consider all requests for the substitution power to be exercised, or any process laid out for how these substitutions are to be considered and resolved. In other words, an individual would approach the CEO as a supplicant. If the CEO chose not to exercise their supervisory power, there would be no way of compelling them to do so.

More broadly, the non-delegable CEO substitution power may have the following adverse impacts:

  • significant backlogs and delays as every single substitution request is funnelled through the bottleneck of the CEO’s office

  • only a small proportion of incorrect decisions are likely to be actually considered by the CEO for substitution

  • minor but still material errors may not be picked up or addressed at an early stage, or at all, and these can spiral into more serious problems for individuals that may take significant time and effort to undo

  • systemic errors or large-scale problems with ADM systems are more likely to slip through the cracks

  • this system provides for review by an official who is removed from day-to-day decision making, with limited experience of working with NDIS participants. The CEO should be the last port of call, not the first.

For these reasons, the Bill should explicitly provide for decisions to be substituted at the officer level. Officers should be able to substitute an automated decision for a correct or preferable decision in all the conventional circumstances that this power is exercised. Among other things, it allows for an error or other substantive concern to be addressed at an early stage, unless this substitution would compromise the integrity of the decision-making process. In other words, substituted decisions should not be less accurate than the automated decision, and the choice to substitute a decision should not undermine human rights principles (for example, a beneficial automated decision

11

Submission 347

29 May 2026

should not be substituted for a detrimental substituted decision unless there are clear and justifiable grounds for doing so).

At the very least, the CEO substitution power should be delegable to NDIS staff – this would necessitate deleting proposed s 202(2A) of the NDIS Act.

Recommendation 5

(c) The Bill should be amended to include a provision enabling decisions to be substituted by officers for a correct or preferable decision, in circumstances where the substitution will not compromise the integrity of the initial decision (i.e. will not be less accurate, or contrary to human rights principles).

(d) Sched 3, Part 2, Item 12 (proposed s 202(2A) of the NDIS Act) should be deleted.

Hidden limitations on internal and external review

The decision to make the CEO substitution power non-delegable is an unusual choice, and the reasoning behind this choice is not obvious from the Explanatory Memorandum. These would be the only powers of the CEO that are explicitly prohibited from delegation under the Act.

Proposed s 59B(11) of the Bill provides that the substitution provision does not limit any other provision of this Act that provides for the review or reconsideration of an administrative action. However, when read in conjunction with the Act, the apparent effect of the non-delegation provision would be to limit individuals’ rights to seek internal merits review of automated decisions. This is because substitution powers are exercisable only by the CEO, and per s 100(5A) of the NDIS Act, the CEO cannot review their own decisions. Further clarification is needed as to the effect of the substitution provision on access to internal review.

Additionally, while not addressed in the text of this Bill, HTI understands that administrative rules proposed to be applied to the new framework and use of the I-CAN system would have the effect of limiting access to external merits review. The Department of Health, Disability and Ageing’s consultation materials on new framework planning indicate that support needs assessments, and consequently the supports to be funded under the NDIS plan and therefore the overall budget, will not in and of itself be a reviewable decision; it will be considered ‘an input into the decision to approve a plan’.30 This would mean that the Administrative Review Tribunal (ART) would no longer be able to directly amend or vary an NDIS plan, and would instead be limited to ordering the NDIA to conduct another assessment.31 In light of the significant impacts of support needs assessments on individuals’ human rights, it would not be acceptable to restrict merits review in this way.

The potential implications of NDIS changes in this Bill and in proposed NDIS rules for internal and external merits review are highly concerning. This points to a reliance on

30 Australian Government, ‘Reviews and appeals under the new ways of planning (from mid-2026)’ (Fact Sheet, Department of Health, Disability and Ageing, January 2026) 1. See discussion in Law Council of Australia, submission to the Department of Health, Disability and Aging, NDIS rules: public consultation on new framework (March 2026) <https://lawcouncil.au/publicassets/58ae601f-572d-f111-94c5-005056acd090/4818%20-%20S%20 %20NDIS%20Rules%20-%20New%20Framework%20Planning.pdf>. 31 Law Council of Australia, submission to the Department of Health, Disability and Aging, NDIS rules: public consultation on new framework (March 2026) <https://lawcouncil.au/publicassets/58ae601f-572d-f111-94c5 005056acd090/4818%20-%20S%20-%20NDIS%20Rules%20-%20New%20Framework%20Planning.pdf>; See also, Guardian reporting confirming that the NDIA had received legal advice to this effect: Kate Lyons, ‘NDIS plans will be computer-generated, with human involvement dramatically cut under sweeping overhaul’ The Guardian, 3 December 2025 < https://www.theguardian.com/australia-news/2025/dec/03/ndis-plans-computer-generated>.

12

Submission 347

29 May 2026

ADM systems to obscure decision-making processes, in order to get around rights of review that would otherwise be available.

In the financial year 2024-25, there was a 77% increase in the number of new cases brought before the ART of people appealing decisions by the NDIA. In 73% of these cases, the NDIA decision under review was changed – which shows that ART merits review is an essential safeguard.32 Rather than addressing the underlying issues that are resulting in an uptick in appeals and the high proportion of decisions being remade, this appears to be being swept under the rug, at the cost of NDIS participants’ access to justice.

Upon the establishment of the ART, which was set up to implement Robodebt Royal Commission Recommendations, the then Attorney-General stated that merits review is ‘critical to Australia’s system of government’.33 Merits review improves the quality of decisions made by primary decision-makers and provides an important accountability check.

For these reasons, the bill should guarantee external and internal merits review rights, including by protecting against reduction in those rights via changes to administrative rules. As a general principle, any limitation on rights of review should be clearly and publicly justified in primary legislation, based on sound principles of law and public policy.34

Recommendation 6

(c) Proposed s 59B(11) of the Bill should be amended to:

o preserve rights of internal and external merits review for decisions made under designated or specified provisions; and

o provide that any administrative decision made or materially influenced by ADM under the Act should be subject to external merits review in circumstances where the decision would be reviewable if a human decision maker had made it.

(d) The NDIA should also set up accessible pathways for seeking internal review of automated decisions made under the Act.

Need for additional safeguards

The Bill includes some welcome safeguards applicable to designated and specified provisions under the Act, in the form of notice and transparency measures. While these safeguards are important, they do not address more fundamental considerations such as accountability for decisions, explanations of decisions, and effective oversight.

The safeguards also do not extend to other forms of automation that the NDIA has determined do not require specific statutory authorisation to adopt. A legislative note at the end of proposed s 59B(1) provides that the Bill is ‘not intended to limit the use of

32 Administrative Review Tribunal, Annual Report 2024-25 (2025, Online) ‘Jurisdictional area caseload’ < https://www.transparency.gov.au/publications/attorney-general-s/administrative-review-tribunal/administrative-review tribunal-annual-report-2024-25/chapter-3%3A-performance/jurisdictional-area-caseloads>; Kate Lyons, ‘NDIS plans will be computer-generated, with human involvement dramatically cut under sweeping overhaul’ The Guardian, 3 December 2025 < https://www.theguardian.com/australia-news/2025/dec/03/ndis-plans-computer-generated>. 33 Parliament of Australia, Parliamentary Debates, House of Representatives, 7 December 2023, 9198 (Mark Dreyfus). 34 See, e.g., Joel Townsend and Emily Singh, ‘The allocation of merits review rights in Australia: anomaly and variation’ Australian Public Law (Blog post) March 2024 < https://www.auspublaw.org/blog/2024/3/the-allocation-of-merits-review rights-in-australia-anomaly-and-variation>.

13

Submission 347

29 May 2026

automation for an administrative action if that action would otherwise not require statutory authorisation in order to be validly taken’.35

This could theoretically include a wide range of administrative actions impacting people. For example, the Explanatory Memorandum mentions proposed s 50A as one such provision. This section provides for the renewal of NDIS plans, which would appear to necessitate some consideration of individual circumstances by a human, such as changes in circumstances, to ensure fairness. HTI queries whether this is an appropriate use of automation, regardless of whether statutory authorisation is required, or whether the NDIA has determined that it does not technically constitute an administrative decision.

The safeguards proposed below should be adopted in the Bill, and extend to all NDIS decisions or administrative action made or materially influenced by ADM, where it may impact people’s legal or human rights – not just those designated or specified under the Act.

Human oversight

Specific protections for human oversight and responsibility should be incorporated into the Bill. Genuine human oversight should be guaranteed when:

  • an administrative decision or action may have a detrimental impact on an individual (for example, rejection of eligibility, reductions in funding)

  • there are discretionary or evaluative judgements involved (as above)

  • there is a risk of error that is capable of being mitigated though human oversight.

Explanations for automated decisions

Proposed s 59E(4) provides a person must be informed of the fact that an administrative action affecting them was automated. In addition to informing people of automation used in administrative actions, when an administrative decision is made affecting a person’s legal rights, the person should be provided with an explanation of the legal and factual basis for the decision, and how to seek review of the decision.

The purpose of the explanation should be to equip a person to exercise rights of review relation to the decision, should they wish to do so. This is necessary to prevent a reversal of the ‘burden of proof’ onto an individual, as occurred in the case of Robodebt, where recipients had no means of understanding how a debt was calculated, and therefore no practical ability to challenge the debt.

There should also be a requirement for affected people to be notified when they have been subject to an ADM decision that has been affected by error, so they can seek redress or take other appropriate action.

Testing and transparency

The Bill should include a requirement for relevant ADM systems to be subject to independent performance testing, before and after deployment. The results of these tests, including error rates, should be published in the NDIA’s annual report, alongside other information outlined in proposed s 59E(7).

35 Explanatory Memorandum, National Disability Insurance Scheme Amendment (Securing the NDIS for Future

Generations) Bill 2026 (Cth) 128.

14

Submission 347

29 May 2026

As recommended by the Robodebt Royal Commission, where possible, the NDIS should release the business rules, algorithms and operating procedures associated with ADM systems, to improve accountability and enable testing by external parties.36

External oversight

The Bill provides criteria for the CEO to comply with when developing the standard operating instrument. For example, proposed s 59E states that the CEO is required to take steps to ensure that the administrative action can be validly taken. However, failing to take these steps does not invalidate decisions, and there is no mechanism for external oversight to ensure that the instrument and the decisions made under it are in accordance with the Act. The Bill should provide for the CEO to report to an independent agency, such as the NDIS Quality and Safeguards Commission, or the Commonwealth Ombudsman, who can assess the contents of the instrument and other administrative instruments related to ADM use, and their practical implementation.

Human rights considerations

The CEO should be required to consider the human rights set out in the Human Rights (Parliamentary Scrutiny) Act 2011 (Cth) when developing the standard operating instrument or any other administrative instrument related to ADM, to ensure that the full range of possible harms and impacts on affected individuals are factored into operational procedures.

Recommendation 7

(c) The Bill should be amended to include additional safeguards for all automation of administrative action under the Act, where automation would have the potential to reduce individuals’ legal or human rights.

(d) These safeguards should include specified requirements for: genuine human oversight; reasons for decisions; notification of ADM errors; testing of ADM systems; public information about the technical workings of ADM systems; external oversight of ADM systems; and consideration of human rights in the development of administrative instruments.

Implications for NDIS culture and staff

As well explored by the Robodebt Royal Commission, humans may have limited motivation, opportunity or power to disrupt highly-prescriptive processes dominated by automated systems.37 This in turn leads to a reduction in administrative and professional knowledge and a loss of legitimacy for government institutions.38 Limiting the discretion and autonomy of staff means that they are less able to advocate for, and meet the needs of NDIS participants, or raise problems up the chain before they become unmanageable.

Additionally, the success of ADM systems depends on effective socio-technical systems – in other words, humans and machines should work well together. This can only occur if the human is properly supported to understand the technical decision making processes they are engaging with (including a realistic understanding of the system’s strengths and limits) and the rules they are administering. Investment in ADM

36 Royal Commission into the Robodebt Scheme, Final Report (July 2023) Recommendation 17.1. 37 See Royal Commission into the Robodebt Scheme, Final Report (July 2023), Chapter 23 ‘Improving the Australian Public Service’. 38 Paul Henman, ‘Digital Social Policy: Past, Present, Future’ Journal of Social Policy (2022) 3, 535-550, 538. See also Simone Casey, ‘Towards digital dole parole: A review of digital self‐service initiatives in Australian employment services’ Australian Journal of Social Issues (2021) 57(1), 114.

15

Submission 347

29 May 2026

systems must be accompanied by an investment in staff – including through dedicated training programs on working with ADM systems. There is no shortcut around the need for qualified, trained and empathetic frontline staff to engage with people with complex needs and their families.

For ADM to operate effectively, the design of ADM systems should also take into account insights drawn from consultation with staff – namely, the users of these systems on a day-to-day basis.

Recommendation 8

(c) The NDIA should invest in training NDIS staff so they can effectively review ADM decisions, with an understanding of ADM technology they work with, the underlying law, and the application of human-centred AI and ADM principles.

(d) The NDIA should consult with NDIS staff before adopting ADM systems that staff will interact with, or rely upon, to make decisions

16