Publication
HHS delegates exclusion authority to CMS
HHS Secretary Robert F. Kennedy Jr. announced that the CMS would exercise exclusion authority to remove “bad actors” from federal healthcare programs “effective immediately.”
Australia | Publication | August 2026
Automated decision-making (ADM) is not a future concept for government; instead, it is already embedded in the way services are delivered, from tax administration to border processing. As agencies move to scale decision-making through automation and AI, the legal and reputational stakes are rising just as quickly. Recent scrutiny of government systems has reinforced a simple point: automation may improve speed and consistency, but it must not displace fundamental public law obligations. Against this backdrop, agencies must ensure that automated processes are not only efficient, but lawful, transparent and fair. This article highlights the key legal considerations for Commonwealth entities seeking to harness automation while maintaining public trust.
ADM systems offer significant benefits for government agencies looking to do more with less. However, they introduce several key risks that must be carefully managed.
Key legal risks for automated administrative decision-making include:
The use of automated decision-making in government will only expand, particularly as artificial intelligence (AI) capabilities mature. For Commonwealth agencies, the key challenge is how to adopt automation responsibly. This requires embedding legal safeguards at every stage of the system lifecycle, including design, deployment and review, while maintaining clear lines of accountability and human oversight. Agencies that get this right will not only manage legal risk more effectively but also strengthen trust in government decision-making in an increasingly automated environment.
The work of government impacts the lives of individuals in a myriad of ways, including delivering services and payments to individuals, assessing and collecting taxes, and granting visas, amongst other things. Although ADM has been in use in the Australian public sector for over 30 years, its use by government agencies is becoming more widespread following recent technological developments. Current applications include automated facial recognition and document verification to process travellers at airports,2 and data-matching and rules-based systems to pre-fill and assess individual tax returns.3
Automating aspects of the administrative decision-making process comes with the promise of more efficient and consistent administrative processes, but also raises a number of risks which must be managed carefully. This article provides a high-level overview of some key issues surrounding the automation of administrative decision-making.
Before considering issues related to the automation of administrative decision-making, it is helpful to set out what kinds of “decisions” are “administrative” in character. First, a “decision” (in the sense used in administrative law) is a decision that is final or operative and determinative of the matter being considered.4 Second, in broad terms, an “administrative” decision is a decision that is made “in the exercise of statutory powers and functions, which is neither legislative nor judicial in character”.5 Administrative decisions are often distinguished from legislative decisions on the basis that “…legislative decisions determine the content of rules of general, usually prospective, application, whereas administrative decisions apply rules of that kind to particular cases.”6
In the context of administrative decision-making, “automation” refers to the use of computer systems to automate some or all of the decision-making process. Automated systems are diverse, ranging from:
The most common type of ADM currently utilised by Commonwealth agencies are simple rules-based expert systems,8 which automatically apply a series of pre-programmed rules written by humans.9 A well-known example of the use of expert systems in recent years is the Robodebt Scheme, which involved using automated data-matching to identify and recover welfare overpayments.
However, ADM can also include specialised systems that use automated tools to forecast and reason, such as machine learning (a form of AI10).
As Justice Melissa Perry stated:
“Governments currently rely primarily on pre-programmed rules-based automated systems, because wholly “autonomous” decision-making systems remain expensive and have yet to be perfected. However, as technology continues to improve and the use of “machine learning” (or algorithms, which learn from patterns in data) is increasingly adopted in the private and public sectors, we can expect machines to be used to make a greater number of qualitative assessments and recommendations, and even evaluative decisions.”11
These systems raise both overlapping and distinct issues in relation to administrative decision-making. In the following section, we outline some of the most common issues Commonwealth agencies may face in using ADM systems for administrative decision-making. This article focusses primarily on forms of ADM where rules are explicitly encoded in the system (expert systems), as opposed to being ‘found’ in training data, as is the case with AI systems based on machine learning.
There are at least five significant issues and risks raised by the automation of administrative decision-making. These are legal authority to automate, bias and discrimination, lack of transparency, privacy issues, and issues around accountability and responsibility. In this section, we outline the nature of each risk, the relevant law, and what agencies and legal practitioners can do to mitigate and manage each risk.
The use of ADM to make administrative decisions is confined to circumstances where the power or authority to do so is provided for in legislation.12 As noted by Justice Melissa Perry, this is because of the principle of legality:
“….When it comes to the State, a positive justification is required for legal action affecting rights and interests…. Equally, if an automated system is being utilised to make part or all of a decision, the use of that system must be authorised by law.
It cannot be assumed that a statutory authority vested in a senior public servant which extends by implication to a properly authorised officer, will also extend to an automated system; nor that authority to delegate to a human decision-maker will permit “delegation” to an automated system. Authority to use such systems should be transparent and express.”13
Agencies must therefore ensure automated systems are designed so the system complies with the legislative authority for the relevant decision, as well as the authority to automate the decision.14
The legislative authorisation of ADM in administrative decision-making is fragmented, with at least 64 separate pieces of Commonwealth legislation authorising the use of ADM.15 This fragmentation has been criticised for creating inconsistency around which decisions should be automated, leaving gaps in safeguards, complicating oversight, and undermining legal certainty and public trust.16 In 2023, the Royal Commission into the Robodebt Scheme recommended that the Commonwealth consider legislative reform to introduce a consistent legal framework for automation in government services,17 a project which has been taken up by the Attorney-General’s Department following a public consultation process.18 The Attorney General’s Department is progressing a whole of government reform agenda aimed at establishing a more consistent statutory and governance framework for ADM across the Commonwealth.
Since the consultation process concluded, there have been related developments.
The Commonwealth Ombudsman has recently recommended that across the whole of the APS, processes should be reviewed to determine whether there are sufficient safeguards covering the use of automated decision-making, signalling an expectation of systematic assurance rather than ad hoc compliance.19
In December 2025, the Australian Government released its National AI Plan (the Plan). The Plan confirmed that existing legal frameworks already apply to AI. The Plan explicitly stated that the government is focused on a consistent legal framework for supporting government use of ADM (which may include AI), in service delivery to the Australian public.20 Importantly, the Plan confirmed that Australia will not introduce standalone AI legislation in the near term, placing greater weight on agencies to ensure existing administrative law, privacy and accountability frameworks are applied rigorously to ADM systems.
Subsequent Ministerial announcements indicate that ADM within federal agencies remains a specific focus of the Government’s AI safety agenda. On 20 July 2026, the Government identified the development of a framework to better regulate the use of ADM within federal agencies as one of its whole-of-government AI safety priorities. The stated objective is to support fair, accurate and transparent government decision-making, including where emerging technologies such as AI are used. This development reinforces the need for Commonwealth agencies to treat ADM governance not merely as a systems-design issue, but as an administrative law, privacy and accountability issue requiring clear legal authority, transparency, assurance and human responsibility.21
Procedural fairness (or natural justice) requires that an administrative decision is free from bias, both actual and apprehended, and that an individual has the right to be heard.22
To demonstrate actual bias, one must show that a decision-maker is “so committed to a conclusion already formed as to be incapable of alteration, whatever evidence or arguments may be presented.”23
The legal test for apprehended bias contains two parts:
Using certain ADM systems may lead to actual or apprehended bias.
Rules based expert systems rely on the rigid application of ‘if this then that’ logic and can therefore be suited to non-discretionary decisions.25 As the Royal Commission into the Robodebt Scheme observed in its Final Report, that scheme “…involved a system of business rules with no ability to move outside of specific and defined action on the basis of the data received. It was extremely rigid; once the rules had been coded and set in place, the system itself would stay in place until the rules were changed by way of human intervention”.26 On the other hand, discretionary decision-making refers to a decision-making process where the decision maker is allowed some latitude as to the choice of the decision to be made, which latitude may be considerable or quite narrow.27 Questions of pre-judgment or bias may arise where an expert system ADM is used to make discretionary decisions (as well as other issues such as invalid or improper exercise of power through the fettering of discretion of the decision-maker).28
The Commonwealth Ombudsman’s Better Practice Guide on Automated Decision Making, published in March 2025, recognises that ‘some data analytics techniques that support ADM such as automatic algorithms have the potential to create personal information with an inherent bias, that is discriminatory or that leads to inaccurate or unjustified results.’29 This is because AI infers patterns from existing data. If the training data contains embedded biases, the models are likely to replicate and reinforce those biases. In general, this type of bias may be mitigated, but not eliminated.
The impact of errors, biases and even unreasonableness in automated administrative decision-making processes is potentially much greater than in respect of decisions made by humans alone. First, unlike human decision-making, many forms of ADM rely on an explicit encoding of rules into a system.30 The process of encoding rules into a system may involve errors, ranging from typographical mistakes to inaccuracies in translating the law into code (such as occurs when the law is oversimplified or misunderstood).31
More fundamentally, there is a risk that “the relatively constrained vocabulary of computer code will not adequately reflect the nuances of statutory provisions”.32 Depending on the type of error, this could result in discrimination against certain groups, as well as errors which could amount to unreasonableness.
Thirdly, due to the consistent application of algorithms and the speed of ADM that allows for mass decision-making, any coding error that influences the outcome of decisions could have widespread effects – as was the case with the Robodebt Scheme. This means the impact of errors and biases in automated administrative decision-making processes is potentially much greater than in respect of decisions made by humans alone.
To minimise the risk of errors, lawyers experienced in statutory interpretation and administrative law should be involved early in the design process to ensure that statutory language is accurately translated into code. Care should be taken to ensure that any practical limitations of coding language or expert system logic do not result in questions being phrased in a biased or leading manner.
The risks of coding or other errors in ADM systems are also compounded by the existence of automation bias. This is the susceptibility of people to defer to and accept the outcomes of automated processes due to a perception that these processes are infallible.33 Automation bias may lead to inadequate review of automated steps in a decision-making process by the decision-maker, and may also dissuade affected individuals from seeking review of erroneous decisions. Accordingly, where ADM is used, there should be a clear path for those affected by decisions to seek review as recommended by the Royal Commission into the Robodebt Scheme,34 and an internal process to ensure decision-makers exercise appropriate judgement about the output of ADM systems.
To satisfy the second element of procedural fairness, the right to be heard, individuals affected by a decision should be notified that ADM has been used and be provided with an opportunity to present their views on the information being used to make a decision.
As noted in the Automated Decision Making – Better Practice Guide:
“Transparency is a key value of administrative law and critical for government accountability. The underlying business rules of an automated system must be readily understandable and information about automated systems should be publicly available. People should be informed when automated systems are being used to make decisions that materially impact on their legal rights or other significant rights.”35
Transparency is essential both to reduce the risk of bias and discrimination, and to ensure that individuals affected by an ADM-supported decision can know that such a system was used, to understand the decision, and to challenge or seek review of the decision. In this respect, the type of system used will impact on transparency considerations.
For example, complex ADM systems involving AI do not rely on a fixed pre-determined set of rules and variables, complicating the task of transparency in decision-making. The manner in which such systems make decisions may not be comprehensible, even to their creators. This may make AI systems inappropriate for certain decision-making purposes.
In contrast, the decision-making process of rule based expert systems should be highly transparent, as an expert system applies a pre-determined set of rules and variables in the same way each time. This should mean that it is possible to trace how a decision was arrived at.36 However, in practice, transparency may be challenging for various reasons, for example:
To ensure an appropriate level of transparency, the design of administrative ADM systems should include clear audit trails and provide reasons to accompany decisions. Those reasons should also include mechanisms for review of the decision and should not be finalised without human oversight.
It is also worth noting the Robodebt Royal Commission’s recommendations that:
A consistent theme across recent reviews is the need for stronger assurance mechanisms, including auditability of systems, independent scrutiny of algorithms, and clear internal governance over ADM outputs. In January 2026, the Office of the Information Commissioner (OAIC) published a report40 of its desktop review of how transparent 23 Commonwealth agencies were about their use of ADM in decision-making processes. The review considered whether a member of the public could find out if an agency was using ADM by searching the agency website.
The OAIC review found that:
The OAIC concluded that none of the agencies demonstrated practices to fully enliven the legislative intention in section 7A of the FOI Act to “…publish a range of information including information about what the agency does and the way it does it, as well as information dealt with or used in the course of its operations, some of which is called operational information.”41
The OAIC indicated it would update Part 13 – Information Publication Scheme of the FOI guidelines so that ADM is expressly included as an example of ‘operational information’.42
These findings signal a gap between legislative requirements and current agency practice.
ADM systems inherently involve the processing of personal information at scale. It is therefore essential to ensure that individuals' data is handled lawfully, ethically, and securely, to mitigate risks of misuse, unauthorised access, or data breaches. This includes considering data collection, storage, processing, and sharing within automated decision-making workflows.
From 10 December 2026, amendments to the Australian Privacy Principles will come into effect. Relevantly, APP entities (which include Commonwealth agencies) will be required to include information about ADM in their privacy policies, where the following criteria are met:43
The information provided must include details about the kind of personal information used in making the decisions, and the kinds of decisions made by the computer program.44
These amendments represent the first regulation of ADM within the Privacy Act framework, requiring agencies to proactively identify relevant ADM processes and ensure their privacy settings, data inputs and disclosures are appropriately documented. The OAIC intends to release guidance to assist entities to comply with the ADM transparency obligations by September 2026, and has sought feedback on an Issues Paper.45
In addition to the specific requirements for automated decisions, the other 13 APPs governing collection, use, disclosure, security, and access to personal information remain fully applicable where an ADM system handles personal information.46 Agencies should ensure that their systems are capable of, and do, comply with these privacy requirements.
The Government’s July 2026 AI consumer safety priorities also confirm that privacy reform remains central to the AI safety agenda. In particular, the Government has indicated that it will consult on a second tranche of privacy reform to strengthen, modernise and simplify Australia’s personal data protection laws, noting that a robust framework for personal data protection is crucial for both conventional and AI-driven services. For agencies using ADM, this underscores that compliance should not be limited to the forthcoming ADM-specific privacy policy disclosure requirements. It should also include a broader assessment of whether data collection, use, retention, security and governance settings remain appropriate for automated and AI-supported decision-making.
Legislative provisions authorising ADM typically deem decisions made by computer programs to be decisions made by the relevant authorised officer.47 Agencies should ensure that their processes reflect this ultimate responsibility, namely that accountability for automated decisions remains with human decision makers, notwithstanding the increasing sophistication of the systems used to support or generate those decisions. Agencies must be aware that an automated output may be incorrect and be ready to review that output if a serious question arises.48
This emphasis on human accountability is consistent with the Government’s broader AI safety priorities, which seek to strengthen protections, increase accountability and support the safe adoption of AI within the Commonwealth Government itself. In practical terms, agencies should ensure that responsibility for ADM systems is not diffused across legal, policy, operational and technology teams in a way that makes it unclear who is accountable for the legality, fairness and accuracy of outcomes.
The use of a computer system to make administrative decisions that are discretionary in nature is more complex. Agencies must ensure that an ADM system is designed in a way to avoid the impermissible fettering of a decision-maker’s discretion.49 In this respect, the Commonwealth Ombudsman recommends that “agencies should avoid automating discretions until they have sought independent external legal advice” which considers whether a discretionary decision must be exercised personally by a decision-maker in the particular context.50
This article has provided a brief overview of some key issues and risks to keep in mind when designing and implementing ADM systems in the context of administrative decision-making.
There are undoubtedly benefits to automating some administrative decision-making. Key benefits may include speed, which may provide immediate certainty to individuals impacted by a decision-making process, as well as greater consistency in decision-making. However, as recent examples illustrate, the nature, speed and scale of the way ADM operates presents as a double-edged sword, in that any error will be replicated on a large scale and has the potential to adversely impact a significant proportion of the population.
While ADM adds novel considerations, care also needs to be taken to ensure continued adherence to fundamental administrative law concepts and public law principles, which apply regardless of whether automation is involved.
This article has indicated some of the main issues raised by ADM, which should be a focus of any review. Taking a lifecycle approach to safeguards can help address issues with automation before significant harm arises, lower the risk of major legal liabilities, and enable a fairer distribution of responsibility for identifying and resolving problems between the administering agency and the recipient of a decision.
Recent developments across government indicate a clear trajectory towards more structured oversight of ADM, including whole-of-government reform, increased regulatory scrutiny of transparency practices, the introduction of ADM-specific privacy obligations, and the Government’s July 2026 commitment to develop a framework to better regulate ADM within federal agencies. This is in addition to existing obligations requiring agencies to publish operational information.
Together, these developments reinforce that agencies must move beyond ad hoc implementation and adopt a systematic, lifecycle approach to the governance of ADM, one that addresses legal authority, explainability, privacy, assurance, review rights and human accountability from the outset.
Senator the Hon Tim Ayres, the Hon Michelle Rowland MP, the Hon Anika Wells MP, the Hon Amanda Rishworth MP, the Hon Dr Andrew Charlton MP and the Hon Dr Andrew Leigh MP, ‘AI consumer safety priorities’ (Media Release, 20 July 2026) AI consumer safety priorities.
Australian Broadcasting Tribunal v Bond (1990) 170 CLR 321, 337 (Mason CJ, Brennan and Deane JJ agreeing). While there are also issues associated with the automation of administrative actions not amounting to “decisions”, they are outside the scope of this article.
Australian Government, National AI Plan 2005 (December 2025), 26.
Anna Huggins, ‘Addressing Disconnection: Automated Decision-Making, Administrative Law and Regulatory Reform’ (2021) 44(3) UNSW Law Journal 1048, 1060-1.
Royal Commission into the Robodebt Scheme (Final Report, July 2023) vol 2, 472 (internal citations omitted).
Commonwealth Ombudsman, “Better Practice Guide: Automated Decision Making” (March 2025), 26.
For further information about the APPs, please refer to the link to the Office of the Australian Information Commissioner’s ‘Australian Privacy Principles Guidelines’ under the ‘Additional Resources’ section below. The Guidelines were updated on 3 October 2025 to reflect changes to the Privacy Act 1988 (Cth) made by the Privacy and Other Legislation Amendment Act 2024 (Cth). Chapter 1 contains high level information about the forthcoming requirement to include information about automated decisions in a privacy policy.
Publication
HHS Secretary Robert F. Kennedy Jr. announced that the CMS would exercise exclusion authority to remove “bad actors” from federal healthcare programs “effective immediately.”
Publication
Now in its seventh year, our annual English Premier League (EPL) football report, Keeping Possession, captures insights from our global sports law team on the key trends and topics shaping the landscape of English football.
Subscribe and stay up to date with the latest legal news, information and events . . .
© Norton Rose Fulbright LLP 2026