The state of the ADM accountability ecosystem in Australia

Where ADM is used to make consequential decisions, it can cause harms of all kinds (Buolamwini and Gebru Reference Buolamwini and Gebru2018; O’Neil Reference O’Neil2016; Yeung Reference Yeung, Karen and Martin2019). This is especially true for “machine learning” and “artificial intelligence” (AI) technologies. “AI” describes computational systems that automate complex information processing tasks through algorithms, statistical methods, and data-driven optimization techniques, including “machine learning.” “Machine learning” is the iterative process of automatically adjusting the parameters of an algorithm through exposure to data to optimize performance on specific tasks. Bender and Hanna have humorously proposed a less grandiose name: SALAMI (Systemic Approaches to Learning Algorithms and Machine Inference; Bender and Hanna Reference Bender and Hanna2025: 5).

Regardless of how we name them, such tools generate life-changing decisions at unprecedented scale and speed. While the government misuses of ADM described above were particularly egregious, many serious harms have come to light over the past 5 years in both the public and private sectors. Misuses of ADM in law enforcement (Angwin et al. Reference Angwin, Larson, Mattu and Kirchner2016; Sentas and Pandolfini Reference Sentas and Pandolfini2017); autonomous vehicle accidents (Henry et al. Reference Henry, Simcock and Snoswell2022); toxic AI chatbots (Fraser and Suzor Reference Fraser and Suzor2025; Roose Reference Roose2024; Xiang Reference Xiang2023); facial recognition and intrusive surveillance in retail settings (Commissioner Initiated Investigation into Bunnings Group Ltd (Privacy) 2024; Jung and Kwon Reference Jung and Kwon2024; Kind Reference Kind2025); denial of financial services to sex workers and sexual health providers (Stardust et al. Reference Stardust, Blunt, Garcia, Lee, D’adamo and Kuo2023); intrusive social scoring (Blakkarly Reference Blakkarly2022); and discriminatory decision-making in education (Hern Reference Hern2020) and finance (Garcia et al. Reference Garcia, Garcia and Rigobon2024), have all made news.

There is a pressing need for scholarship on how to apply existing laws to ADM harms. The Australian government recently abandoned a proposal to introduce “mandatory guardrails” for high-impact AI (Safe and Responsible AI in Australia: Proposals Paper for Introducing Mandatory Guardrails for AI in High-Risk Settings 2024), in favor of a National AI Plan which explicitly states that “established laws remain the foundation for addressing and mitigating AI-related risks” (National AI Plan 2025). Even if the appetite for regulation were stronger, emerging AI and ADM regulation has generally been constrained by a “risk-based” approach and only applies to a relatively narrow range of high-risk applications (Mahler Reference Mahler2020; Schuett Reference Schuett2023).

Existing law therefore has an important role to play in regulating ADM. But the application of existing law to new technologies is uncertain (Bennett Moses Reference Bennett Moses2011; Brownsword Reference Brownsword, Brownsword and Yeung2008; Collingridge Reference Collingridge1980). That uncertainty is particularly pronounced for ADM technologies which have developed very quickly over the past decade. Absent regulation, law develops through litigation – including “public interest litigation.”

A precise definition of public interest litigation is elusive (Marshall and Hale Reference Marshall and Hale2014; McGrath Reference McGrath2008; Oshlack v. Richmond River Council 1998b; Taylor Reference Taylor2020). One of its key features, according to the Australian Law Reform Commission, is that it resolves a question of law, develops the law, and addresses matters in which the broader public has an interest (Costs Shifting – Who Pays for Litigation? 1995: pt. 13.2). Ramsen and Gledhill identify four main features of “strategic litigation”:

1. 1. it seeks outcomes with a long-term impact, going beyond the origins of the claimant’s complaint;

2. 2. it is adaptable to a range of purposes;

3. 3. its objectives are multi-faceted and go beyond creating effects within the court system; and

4. 4. it views “litigation” broadly, to include tribunals and international mechanisms of redress (Ramsden and Gledhill Reference Ramsden and Gledhill2019).

It is generally recognized that public interest litigation is an important vehicle for the protection of the rights of marginalized or disadvantaged groups, or for driving social or legal change in the absence of political will (Durbach et al. Reference Durbach, McNamara, Rice and Rix2013). Class actions, which were introduced into Australia in 1992 under Part IVA of the Federal Court of Australia Act 1976 (Cth) (with state regimes also allowing class actions in state courts), provide one avenue for groups to pool resources in relation to shared harms.

There are numerous stakeholders in public interest litigation, including ADM developers and deployers, individuals and communities affected by ADM, litigants (parties in a suit), litigators (their lawyers), civil society organizations, the state, and the general public. Although the public and litigants are the most impacted by the outcome, it is not necessarily these groups who drive the process. Scholars have explored differences between pro bono, cause, and movement lawyering – with differences mostly arising from the role played by lawyers relative to other stakeholders in litigation (Meeusen Reference Meeusen2026). Our interviews indicated that, in Australia, public interest litigation is often propelled by professionals: consumer groups and lawyers who have already built case theories, who are looking to “recruit” litigants to put theory into practice. This fits the profile of “cause” lawyering, though many participants emphasized the importance of ensuring more voice for other stakeholders, in line with “movement” lawyering.

Public interest litigation in Australia faces several well-known barriers. Costs of litigation may be prohibitive (Edwards Reference Edwards1999; Ginnivan Reference Ginnivan2016). Losing parties may, in addition to their own legal costs, be required to pay the costs of counterparties under “adverse costs orders” (Costs Shifting – Who Pays for Litigation? 1995). Finally, those most motivated to bring claims may be obstructed by a lack of “standing” or eligibility (Cane Reference Cane1999). Even so, public interest litigation has influenced some of Australia’s most significant socio-political issues, including the recognition of Indigenous land rights (Mabo v. Queensland (No 2) 1992), environmental protection (Tasmania Dams Case 1983), asylum seeker detention (Al-Kateb v. Godwin 2004), and same-sex marriage (Commonwealth of Australia v. Australian Capital Territory 2013).

By contrast, litigation has yet to produce significant legal clarification with respect to ADM, although Robodebt litigation certainly precipitated political action. Few significant cases dealing with ADM have come before the court, and the success of these cases has been mixed. The Pintarich case – where the majority held that a computer-generated letter from the tax office was not a “decision” capable of judicial review because it was not produced by a human mental process of deliberation – generated more confusion than clarity, despite Justice Kerr’s powerful dissent. In the Robodebt litigation, an action for judicial review led to consent orders stating that the government’s approach to raising automated welfare debts (comparing actual monthly income to average monthly income) was irrational (Amato 2019). Consent orders do not, however, produce legal precedent.

As for litigation in private law, a subsequent Robodebt suit for negligence and unjust enrichment concluded in a settlement, which also does not produce binding precedent (Prygodicz v. Commonwealth of Australia (No 2) 2021). More promising, the Australian Competition and Consumer Commission’s consumer lawsuit against the travel website, Trivago, successfully applied the law of misleading and deceptive conduct in relation to claims about the operation of a ranking algorithm (Australian Competition and Consumer Commission (ACCC) v. Trivago NV ). The case did not settle any particularly thorny issue of law, though it did establish a helpful roadmap for leading expert evidence about algorithmic systems and behavioral science (Henry et al. Reference Henry, Simcock and Snoswell2022). There is much more scope for public interest litigation to tackle the problem of ADM accountability.

Aims and method

This project uses applied qualitative social research to investigate the challenges, opportunities, barriers, and enablers for effective public interest litigation for ADM accountability. Applied social research addresses problems raised by organizations, governments, or businesses and is designed to solve immediate and persistent problems (Hall Reference HHall2008). The project was conducted within the Centre for Automated Decision-Making and Society, which was specifically funded by the Australian Research Council as a 7-year “Centre of Excellence” initiative, bringing together universities, industry, government, and community to investigate the development of responsible, ethical, and inclusive ADM. The goal of this project was to achieve real-world change, bringing together stakeholders who encounter and challenge ADM systems in their daily practice, and synthesizing their insights into practical recommendations that can be employed by advocates, communities, and practitioners.

To do so, we recruited and interviewed 20 candidates with experience in public-interest litigation, advocacy, technology law, and policy. We obtained ethics approval for the study from our faculty ethics committee before commencing. Participants included barristers, solicitors, policy advisors, academics, and consumer advocates from both public and private institutions. We used our own networks, based on our previous experience as lawyers and law reform advocates, and developed during our research fellowships at the Centre for Automated Decision-Making and Society to recruit participants initially, and then used snowball sampling based on participant recommendations. This sample size is typical for qualitative research, which seeks to elicit rich, in-depth, textural data that offers insights into the world, rather than generating population-level or generalizable data. Our sample size was sufficient for us to reach the saturation point, whereby new data were not yielding additional insights (Hennink and Kaiser Reference Hennink and Kaiser2019). It is also worth noting that the Australian public interest litigation and digital rights ecosystem is small compared with jurisdictions like the US or UK.

We conducted and recorded one-hour semi-structured interviews and gave participants the opportunity to review and edit transcripts for clarity and confidentiality. We used an outline of questions for the interview, but allowed conversation to flow naturally based on participants’ answers, interests, and expertise. The interviews were specifically related to public interest litigation in the context of ADM. ADM provides a unique challenge for public interest litigation, given the speed of development of ADM technologies, their application to large domains of economic and social life, the lack of accountability of governments and corporations for harmful ADM, uncertainty about how the law applies, and the sheer scale of actual and potential harm.

We then conducted qualitative thematic analysis of the data (Braun and Clarke Reference Braun and Clarke2006), identifying, analyzing, and interpreting patterns of meaning (themes). Our first set of codes was organized around the structure of the interviews. These included opportunities for public interest litigation (identifying relevant cases and areas of practice); barriers (which ranged from aggressive litigation and system opacity to lack of access to justice and systemic harms); and enablers (including funding, aggregating institutions, law reform, legal cultures, referral networks, and monitoring). We then developed further conceptual codes, taking an iterative approach to coding that drew from our contemporaneous file notes, reflections, and memos on the significance of the data (Strauss and Corbin Reference Strauss and Corbin1998). These codes included doctrines, power, and movement lawyering – featuring data about who drives litigation, the limits of the law, and the role of lawyers – as well as retrofitting, aligning stakeholder interests, and the transparency-accountability lifecycle.

There are several limitations to our study. Participants were mostly professionals (lawyers, members of civil society organizations, academics).Footnote ¹ Our understanding of the experience of litigants was thus mediated by these professional advocates, notwithstanding their deep knowledge of the issues. A key direction for future research is prioritizing perspectives from impacted communities who are experts in their own experience and can offer bottom-up insights into how ADM operates. Only one participant was an administrative law expert. Consequently, although our study was partly motivated by the Robodebt cases (of which one was a judicial review action), interviews did not generally produce deep dives into ADM and administrative law.

The data are most relevant to the Australian legal context, where doctrines like administrative law and discrimination are idiosyncratic, and where the digital rights advocacy ecosystem is relatively small. Other jurisdictions have different institutional parameters and legislative contexts (for example, in relation to costs waivers). However, our findings around tactics of retrofitting may be relevant to other common law jurisdictions, especially those that have not introduced strong AI regulation.

Related scholarship

This article builds on and contributes to the scholarship of how to develop and apply existing laws (as well as new regulations) to ADM. Scholars have analyzed both conceptual and practical challenges of doing so. Important work, for instance, explores whether and how foundational concepts such as “accountability,” “public interest,” or “explanation” apply to decisions made or mediated by ADM (Cobbe Reference Cobbe2019; Cohen and Suzor Reference Cohen and Suzor2024; Miller Reference Miller2019; Singh et al. Reference Singh, Cobbe and Norval2018; Wachter et al. Reference Wachter, Mittelstadt and Russell2018). Other work documents the ways in which the opacity, complexity, and inscrutability of AI systems obstruct accountability (Burrell Reference Burrell2016; Pasquale Reference Pasquale2015; Selbst and Barocas Reference Selbst and Barocas2018). A growing literature critically evaluates the effectiveness of new regulations such as Europe’s AI Act (Engler and Renda Reference Engler and Renda2022; Fraser et al. Reference Fraser, Parker, Haines, Bello Y Villarino and Weatherall2024; Hacker and Holweg Reference Hacker and Holweg2025; Mahler Reference Mahler2020; Schuett Reference Schuett2023; Veale and Borgesius Reference Veale and Borgesius2021), and existing doctrines – such as equity (Balkin Reference Balkin2015; Zhou et al. Reference Zhou, Lu, Chen, Zhu, Xu, Xing and Harrer2024), discrimination law (Barocas and Selbst Reference Barocas and Selbst2016; Mann and Matzner Reference Mann and Matzner2019; Sheard Reference Sheard2022), tort (Buiten et al. Reference Buiten, de Streel and Peitz2021; Fraser and Suzor Reference Fraser and Suzor2025; Karnow Reference Karnow, Calo, Froomkin and Kerr2016; Selbst Reference Selbst2020; Yoshikawa Reference Yoshikawa2018), and administrative law (Cobbe Reference Cobbe2019; McGurk and Tomlinson Reference McGurk and Tomlinson2025; Ng Reference Ng2025) – in addressing harm and injustice from ADM.

We extend this literature by exploring the practical and tactical implications, for litigants and litigators, of the strengths and weaknesses of different doctrines. In doing so, we connect the conceptual literature to more practically oriented scholarship. Private law scholars have delved into challenges and opportunities of developing and leading evidence and civil procedure in cases involving AI and ADM (Caruso et al. Reference Caruso, Legg and Phoustanis2019; Henry et al. Reference Henry, Simcock and Snoswell2022; Llorca et al. Reference Llorca, Charisi, Hamon, Sánchez and Gómez2023). Scholars have highlighted institutional practices and attitudes in government that obfuscate ADM harms and obstruct accountability (Carney Reference Carney2019; Henman Reference Henman, Hertogh, Kirkham, Thomas and Tomlinson2021, Reference Henman2025). And public law scholarship has tackled the challenge of designing reviewable ADM systems (Cobbe et al. Reference Cobbe, Lee and Singh2021), evidencing public law errors in ADM in judicial review (Tomlinson et al. Reference Tomlinson, Sheridan and Harkens2020) and using data science to formulate new approaches to reviewing the rationality of ADM (Simcock Reference Simcock2026).

We also align with literature theorizing strategies and tactics for public interest litigation such as “winning through losing” (NeJaime Reference NeJaime2010), the relationship between law and organizing (Cummings Reference Cummings2009), and the role of public interest litigation in raising awareness and generating political pressure (Harlow and Rawlings Reference Harlow and Rawlings2013).

Our contribution to that practical literature is not only to bring theories of litigation to bear on the issue of ADM accountability, but also, through our qualitative method, to add granular detail on the practical craft of public interest litigation in the Australian context. Rather than doing historical, political, or doctrinal analysis of cases and movements (like NeJaime or Cummings), our findings come from thematic analysis of our conversations with Australian domain experts. For instance, we enrich Tomlinson et al.’s prognostications about practices of fact-finding and expert evidence likely to evolve to deal with ADM (Tomlinson et al. Reference Tomlinson, Sheridan and Harkens2020) with details about how participants approach decisions regarding cost, resources, and politics. Our approach thus contributes rich insights about how power, politics, and practice affect ADM and public interest litigation in Australia.

Just one tool in a larger toolkit for justice

Almost all participants agreed that public interest litigation has some role to play in ADM governance, though their levels of enthusiasm varied. Participants with a background in litigation tended to be the most positive. For example, Lizzie O’Shea, a solicitor specializing in public interest litigation, working in a large Australian plaintiff law firm, and the chair of advocacy organization, Digital Rights Watch, said:

The framing of litigation as a kind of private regulation (echoed by a number of participants) represents an optimistic view: that individuals enforcing their rights in court using existing laws could exert a broader regulatory effect. This view aligns with a body of regulatory theory literature with a law and economics flavor, which views litigation as a form of user-pays, private regulation which may be more cost-effective and efficient than regulation under certain conditions (Luff Reference Luff2011; Posner Reference Posner2010). One such condition is when change outpaces law-making – and that is very much characteristic of ADM.

Participants with other backgrounds were less enthusiastic. Academics, such as Professor David Lindsay of the University of Technology SydneyTS and Professor Lyria Bennet-Moses of the University of New South Wales, were interested in “zooming out,” and critically evaluating the whole regulatory architecture for governing ADM. As Lindsay put it,

Lindsay and others saw litigation as capable only of achieving “incremental changes”; with regulatory intervention needed for more systematic or significant reform.

Informed by his experience of government decision-making, Professor Terry Carney, a longtime member of the Administrative Appeals Tribunal and an expert in social security and administrative law, was pessimistic for other reasons. He pointed out that litigation in relation to government decision-making is limited, as courts are only permitted to review the legality, rather than the merits, of decisions of the executive branch of government. He suggested that the development of better means to manage merits review of administrative action would permit accountability across a much wider range of concerns. This is perhaps a good reason to understand public interest litigation more broadly, to encompass this kind of administrative action, as well as lawsuits in court (Ramsden and Gledhill Reference Ramsden and Gledhill2019).

Nonetheless, while many participants preferred other avenues for ADM accountability, very few totally dismissed public interest litigation. Vicki Sentas, a legal scholar, warned about the potential for litigation to have “perverse effects” – prompting police, for example, to simply tweak their ADM systems to cement, rather than remedy, injustice. Despite her reservations, she remained heavily involved in public interest litigation through coordinating the Police Powers Clinic in collaboration with UNSW and Redfern Legal Centre. Most emblematic of the less enthusiastic participants was Kate Bower, who at the time of the interviews worked at the consumer peak body, CHOICE, and who is now working at the Office of the Australian Information Commissioner. For her, litigation was “the last stop” when all other forms of advocacy fail. For another public interest litigator turned legal scholar who preferred to remain anonymous, litigation was something that was needed “in the meantime” while waiting for more systematic regulation. These participants perceived litigation to be less appealing than other forms of advocacy and regulation, but still acknowledged that it had a place within a broader ecosystem of accountability.

From law reform to systemic change

If public interest litigation does have some role in promoting ADM accountability, what is that role? Our participants presented a range of views on the goals of public interest litigation which were broadly consistent with Ramsden and Gledhill’s pluralistic characterization of “strategic” litigation (Ramsden and Gledhill Reference Ramsden and Gledhill2019). Min Guo, a barrister who represented the plaintiffs in the Robodebt class action, summed up the goal of public interest litigation as “systemic change,” and this view was common among participants. Interestingly, only one participant included direct pro bono work – lawyers helping individuals or organizations gratis without necessarily adopting a broader political framing (Meeusen Reference Meeusen2026) – in their descriptions of public interest litigation. This may have been a result of participant selection. That participant was the only lawyer from a large commercial firm (as opposed to a plaintiff firm). In any case, participants presented different overall strategies for achieving such change, and our analysis identified three main approaches: asserting the law, developing the law, and advocating change by combining litigation with other forms of advocacy.

One basic strategy is to use public interest litigation to remind ADM developers and deployers that they are not above the law. Professor Ed Santow, a former head of the Public Interest Advocacy Centre (PIAC; now the Justice & Equity Centre), former Australian Human Rights Commissioner, and now a Professor at UTS, said of public interest litigation:

Michael Rivette, a senior barrister, saw legal judgments in litigation as setting new “boundaries” in companies’ appreciation of risk. Companies recognize that if they cross that boundary, they are in a “risk area,” he said. The strategy in such cases may therefore be to achieve systemic change through a change in perception of the law rather than a dramatic change in the law itself.

Another strategy was to develop the law in the public interest through the enforcement of a private interest. Here is how Min Guo described this strategy:

This strategy fits within a broader theme we observed that public interest litigation is an exercise in aligning interests and incentives: in this case, aligning private interests in disputes with broader public interests.

In other circumstances, the goal might be to draw attention to a problem with existing law and policy, in order to prompt changes in policy and legislation. Describing the work of the PIAC, Principal Solicitor (and now CEO of Redfern Legal Centre) Camilla Pandolfini said:

Where the goal is systemic change, public interest litigation works best when it is aligned with a broader strategy of activism. When public interest litigation is integrated into strategic advocacy, its success need not depend on “winning” court cases. Lizzie O’Shea explained that even when public interest litigation “doesn’t work, it can be really effective.” She gave the example of a suit about harassment of staff and patients at the front of the Fertility Control Clinic in Melbourne. The Clinic sued the city council for failing to prevent that harassment. She described the outcome in these terms:

What O’Shea describes aligns closely with what NeJaime describes as “winning through losing” – in this instance, using the litigation loss to galvanize and mobilize activists, and appeal to the public and other non-judicial actors for change (NeJaime Reference NeJaime2010).

Several participants took similar views, with Min Guo suggesting that this strategy amounts to demonstrating to the public that “The law is not on our side, therefore the law needs to be changed.” They explained that galvanizing public opinion in this way requires a combined approach to advocacy integrating legal action, media action, policy advocacy, and awareness raising. These reflections are consistent with research by Newhouse and others (who have combined litigation with coronial inquests, complaints, and education) suggesting public interest litigation is “most effective when conducted in conjunction with public advocacy within a multi-faceted campaign for change” (Newhouse et al. Reference Newhouse, Dozer, Janssen and MacGregor2023).

Embedding litigation in advocacy and movement-building

Participants explained that the success of the “combined” approach to public interest litigation described above depends on it being deeply connected to social movements, broad networks, and communities. Some explained their approach to building these communities. Rivette, for example, spoke about a “brains trust” which included both academics and lawyers who could help him to keep abreast of international developments in privacy law which he might potentially import. “[T]here’s a great depth of knowledge,” he said, “and I see myself as a compiler of that.”

A human rights lawyer who worked with Indigenous communities favored an approach to networking and knowledge-sharing that was still more structured and forward-looking:

By developing theories of how the law could develop to deal with likely scenarios, litigators then need only wait for (or may even actively seek) a client whose circumstances are sufficiently close to the theory. They can then act quickly, decisively, and with the confidence that their approach – which must address the complex and cross-disciplinary nature of ADM systems – is supported by a rigorous and collaborative process of analysis and peer review.

Retrofitting old laws to new harms

Many of our participants recognized that our existing laws do not neatly “fit” the types of wrongs and harms emerging from the increasing use of ADM in consequential contexts. This is particularly the case in Australia, which lacks a federal Bill of Rights. Some states have charters of rights (bringing more potential to challenge ADM harms), but generally litigants must rely on an uneven patchwork of rights (Brennan et al., Reference Brennan2009). Where harms are not specifically covered by legislation, litigants have to rely on adapting common law or old legislative remedies to new circumstances. Min Guo described this as a process of “retrofitting” existing laws. He explained the challenge of legal retrofitting:

An experienced litigator from a legal advocacy organization gave several examples of legal retrofitting in non-ADM contexts. In one case, he used property law to protect the interest of the individual in preventing the arbitrary government confiscation of phones in immigration detention:

He thought that such an approach – falling back on basic flexible, common law doctrines – would fit well for the new kinds of harms associated with ADM.

All cases in common law jurisdictions to some extent involve applying old law to new circumstances, but the need for retrofitting is especially salient for ADM. It is a fast-moving technology. It causes new kinds of harms including automated denial of access to critical private and public services, collective harms like automated bias; and harms which do not fit usual patterns of responsibility, such as autonomous vehicle accidents.

Choosing the right action, client, and remedy

Retrofitting is not simply about making creative arguments in court. Participants were highly tactical about formulating and bringing cases. One of the first choices that litigators face when formulating a claim about ADM is to choose a cause of action. In Lizzie O’Shea’s words: “what kind of harm are we going to seek to remedy and what does that remedy look like?” These constraints are not merely doctrinal (what harms can the law remedy?) but also practical. Public interest litigators can be tactical about which clients and cases to take on – as Min Guo put it, “Is it a set of facts that actually lends itself nicely to existing law?” This question matters when litigators and activists have scarce resources to apply toward their strategic goals. One community litigator explained, “We don’t have the resources to run many cases. So the cases that we take on must have a strategic focus.”

Participants also pointed out a range of other tactical questions: including available remedies and the amount of work needed to update the law. It makes sense for litigators to retrofit causes of action that require less legal and conceptual heavy lifting. Where litigants have already had success with a particular doctrine, it will generally be efficient to continue to use (and further develop) that cause of action. Megan Richardson, a privacy and technology law scholar, spoke of her long-running interest in “how you can use existing causes of action effectively. How you can develop them, create them, move them along, and achieve social outcomes that are desirable?” Michael Rivette observed that there are some causes of action that “seem to have more flexibility built into them” and gave the example of breach of confidence: “in privacy litigation I will always look for a breach of confidence, because we’ve already developed that. It’s a really powerful cause.”

Indeed, litigators can actively work toward this kind of efficiency by strategically retrofitting doctrines over the course of several lawsuits for different clients. Rivette’s approach to developing the law of breach of confidence over 14 years was, he said, conscious and targeted: “me and many others knitting the tapestry together.” He spoke of a series of privacy and breach of confidence cases in which the remedies for privacy infringements expanded (Giller v. Procopets 2008; Jane Doe v. Australian Broadcasting Corporation 2007):

These cases changed the landscape by establishing that the equitable action for breach of confidence could apply to privacy harms, and that the threshold for damages for such a breach could be lowered to “mere distress,” rather than more tangible harms.

Bolt-ons and aligning incentives

A key dimension of retrofitting is aligning incentives. Public interest litigators are working in imperfect circumstances. Pragmatic litigators, we found, will cannily try to align private interests (litigants, funders, and firms) with the public interest. To this end, litigation must do more than merely clarify the application of the law to new circumstances. The public interest dimension comes from seeking some outcome broader than the resolution of a particular dispute. We presumed there would be tensions between the public interest and the immediate interests of litigants in swift and cost-effective resolution. Our participants highlighted, however, that many litigants have the conviction to pursue more systematic change.

Participants were conscious not only of aligning claimants’ incentives with the public interest, but also the incentives of other actors, including funders. Litigators pointed out that it was tactically sound to pursue, where possible, causes of action which provide monetary damages as a remedy for ADM harm. A public-interest-litigator-turned-legal-scholar explained:

In other words, it will make sense wherever possible to include in public interest claims causes of action for which monetary remedies are available.

Doctrines such as negligence, unjust enrichment, breach of confidence, and other torts and equitable actions, participants explained, are all appealing for that reason. The doctrine of negligence, for example, has three things going for it: the remedy is monetary; it has a history of incremental adaptation to new technology; and it is not necessary to advert to the mental state of the defendant (e.g. intent) in relation to the harm. On the other hand, as Carney pointed out, negligence may be difficult to establish in relation to government ADM, because courts are reluctant to recognize duties of care that interfere with existing statutory duties. Equitable doctrines may recognize and compensate for a wider range of injuries than tort, Rivette explained. The recent reform to privacy law (allowing individual actions for serious invasion of privacy) may also generate incentives for litigators to focus on the use of data in ADM (Privacy and Other Legislation Amendment Act 2024).

Participants were generally not overly worried about involving commercial actors in public interest advocacy. They instead sought alignment between those actors’ business interests and the public interest. For instance, Lizzie O’Shea pointed out that enterprising litigation funders may see the funding of a test case, even fairly creatively framed, as an investment in a future revenue stream: “I think funders would be willing to be a bit experimental with some of these kinds of things, because it’s quite clearly a developing field.” If an action is successful and generates a sufficiently important legal development, funders may then have the opportunity to fund (and profit from) future cases cut from the same cloth. The same logic applies to commercial firms, and even barristers. Actors with commercial interests, even interests in taking portions of damages awarded, were generally seen as enablers of effective action. That view is consistent with research which finds that US legal professionals’ motivations for public interest work are varied, and mercenary motivations do not necessarily compromise public interest goals (Cummings and Sandefur Reference Cummings and Sandefur2013).

In formulating claims, remedies matter as much as conceptual and doctrinal potential. To address harms from ADM in government, Lizzie O’Shea pointed out that the “sweet spot” is the intersection between private and administrative law – for example in relation to government IT procurement. That view is consistent with work by McGurk and Tomlinson, which emphasizes the role of procurement law alongside administrative remedies in dealing with government ADM (McGurk and Tomlinson Reference McGurk and Tomlinson2025). Incentives will be better aligned – especially for funders – in suits about government use of ADM when monetary remedies are available – rather than cases where only non-monetary administrative remedies such as declarations or injunctions are available. Robodebt litigation matched that pattern: a judicial review suit led the government to stop using the Robodebt ADM system (Amato 2019), but it was the lawsuit in negligence and unjust enrichment, with its potential for large sums in damages, that made a class action viable, and (as Min Guo pointed out) motivated the government to compensate victims in the settlement (Prygodicz v. Commonwealth of Australia (No 2) 2021).

Another key part of retrofitting includes what we call “bolt-ons” – adding broader public interest matters to otherwise private lawsuits. As Megan Richardson explained, an amicus curiae brief may permit a court to address public interest matters in the context of more narrowly framed lawsuits. She gave the example of the amicus brief by Michael Rivette for the well-known Grubb case (Privacy Commissioner v. Telstra 2017). That case addressed the definition of personal information under the Privacy Act 1988. Richardson said:

In this instance, the amicus brief permitted the court to consider a wider range of public interest issues relating to the definition and regulation of personal information.

Michael Rivette described another tactic he used to expand the scope of lawsuits – which we also think of as a bolt-on tactic. He would include more creative and untested legal claims alongside more conventional, established claims. He explained:

This bolt-on tactic aligns the client’s interest with the public interest. Adding a less established claim in effect diversifies the prospects of success. At the same time, it generates opportunities for systemic retrofitting that create a foothold for future suits. It also represents a practical pathway for private practitioners to realize a widely held aspiration among lawyers: to use “ordinary” commercial practice to promote justice and positive systemic change (Johanson Reference Johanson2020; Sharpless Reference Sharpless2012).

Problematic laws, institutions, and risk-averse legal culture

Attempts to retrofit old laws for new circumstances frequently expose the problems with existing laws and institutions. Australian legal systems carry ongoing legacies of colonialism, racism, sexism, homophobia, and other injustices (Behrendt et al. Reference Behrendt, Cunneen and Libesman2008). Even laws intended to redress some of these harms are limited. For instance, participants viewed anti-discrimination law, which might appear well adapted to address unfair ADM and algorithmic bias (Barocas and Selbst Reference Barocas and Selbst2016), pessimistically.

Discrimination law is limited by a number of problems. Courts do not tend to award large sums in damages, so litigation funding lacks incentives. It lacks the flexibility of doctrines like negligence. And it is based on an outdated model that sees discrimination as an issue of individual bias or prejudice rather than a systemic structure. A litigator involved in movement work with Indigenous communities explained, memorably: “we find it easier to prove that a hospital’s been medically negligent than they have been racist, even if the negligence was based on prejudice.” While negligence is unconcerned with the defendant’s state of mind, discrimination law has difficulties when discrimination arises from something other than bigoted intent.

Direct discrimination does not currently address algorithmic bias based on digital proxies – themselves derived from data reflecting historical injustice (Mann and Matzner Reference Mann and Matzner2019). In our interview, technology professor Lyria Bennet-Moses explained how existing law relies on outdated socio-technical assumptions:

If intent is already difficult to prove, things become further complicated when ADM systems encode intersecting forms of discrimination (Crenshaw Reference Crenshaw and Fineman2013). As tech law scholar Jake Goldenfein put it, “Your data is operationalised in a way that is not intelligible to humans.” There is considerable uncertainty about how the law applies to discrimination on the basis of an algorithmic profile, which is abstracted from protected characteristics such as sex and race (Wachter Reference Wachter2020).

The assumptions of a straightforward, intentional decision motivated by conscious animus no longer work with ADM. This problem is compounded by what Lauren Perry, a Policy and Projects Manager for Responsible Technology at the UTS Human Technology Institute, called “technology deference”:

She recalled multiple instances where the person “on the other end of the phone was saying, well, you have to trust us, because … the system has told us x, y, z.” Automation bias (the tendency to assume that automation removes human “errors” such as discrimination) risks legitimizing decisions that (literally) automatically reenact past injustice.

The doctrine of indirect discrimination seems a promising avenue for addressing automated inequality and injustice. But it too faces significant obstacles. In State and Federal law, generally, indirect discrimination occurs when a person imposes a requirement, condition, or practice that has, or is likely to have, the effect of disadvantaging persons with a protected attribute, and that is not reasonable (see e.g. Equal Opportunity Act 2010 (VIC), s 9(1)(a)). However, the standard of unreasonableness is uncertain with respect to ADM. An academic and litigator specializing in discrimination reflected that judges have “really shied away from making judge-made law” in the area of indirect discrimination. So even where lawyers think the law could be applied, they may not take that avenue because they anticipate “there would be problems, in practice … because judges don’t want to assess if something’s reasonable or not, in this context.” This comment is consistent with scholarly analysis on the challenges of applying indirect discrimination to ADM decisions (Sheard Reference Sheard2022).

This points to a broader problem. Public interest litigation depends on legal culture – the attitudes of lawyers and judges to legal creativity and risk. Adapting the law to new technologies needs a legal profession prepared to apply the law flexibly.

Participants expressed mixed views about the risk appetite of Australian lawyers. Lizzie O’Shea mused about “whether you could convince a barrister to be experimental enough with their advice or be open-minded to thinking about” innovative retrofitting style-claims with respect to ADM. A public interest litigator, she decided, “I feel like there’s enough resources at the bar. They’re not all people who are averse to technology. There are people you can find who will work on these things with you, I think.”

As for the bench, Jake Goldenfein worried that the Federal Court, with a “very black letter” orientation, might not be sufficiently open to new scholarly ideas regarding the development of privacy law:

By contrast, Michael Rivette suggested that other doctrines, such as breach of confidence and negligence, seemed to be associated with a more flexible judicial culture; open to more scholarly or creative arguments, especially since the appointment of Justice Edelman – a former legal academic – to the High Court of Australia.Footnote ² Taken together, participants’ comments suggest that enough of the profession is curious and public-minded enough that they could be powerful advocates for clients harmed by ADM, but that community building strategies described above will play an important role in developing creative legal cultures.

Heavy burdens of litigation limit access to justice

How the law develops in respect to ADM is further influenced by who can afford to bring legal actions – financially, practically, and emotionally. Issues of concern to well-funded litigants will tend to dominate. As David Lindsay pointed out, privacy rights have expanded significantly through court action, especially in the UK, because of celebrity actions.

But people already experiencing marginalization will face additional challenges in bringing public interest litigation, especially in the uncertain area of automated harms. As Samantha Floreani, then program lead at Digital Rights Watch (and now writing a PhD on “rent tech”), put it,

The burdens of litigation may be especially heavy for people in this situation. Vicki Sentas summed up the problem in this way: “There’s a lot of complexity to finding the right litigants and ensuring it’s safe for them as well. And most of the time, it’s not safe.” Camilla Pandolfini expanded on the same sentiment:

Litigation is extremely stressful, with huge time demands. As Lizzie O’Shea put it, “It takes over your life.” Participants further mentioned the risk to litigants of reputational harm, including what Min Guo referred to as the airing of “dirty laundry”: a particularly unappealing prospect for litigants who have already been impacted by the criminal justice system.

A partner at a major commercial law firm, with extensive experience in pro bono work, felt that these burdens block access to justice for people already likely to be experiencing marginalization. They were therefore very pessimistic about public interest litigation:

Their statement evokes the adage “the process is the punishment,” which describes how the experience of going through criminal proceedings in the US can be so taxing, hurried, and unfair as to be a punishment by itself (Feeley Reference Feeley1979). It also evokes feminist scholarship that draws attention to the ways sexual assault survivors are frequently re-traumatized by the criminal justice system (Graycar and Morgan Reference Graycar and Morgan2002), and in particular how Indigenous women are disbelieved as witnesses (McQuire Reference McQuire2024) and punished for seeking assistance (Watego et al. Reference Watego, Macoun, Singh and Strakosch2021).

Entrenching structural harms

Alongside rigid legal cultures and access barriers, there is a risk that public interest litigation cements broader structural harms, despite best intentions. Where litigation is embedded in a broader advocacy strategy around issues with political momentum, its prospects of success are generally better. This creates a subtle incentive to “fit within the dominant political paradigm,” as David Lindsay put it. This may entail prioritizing concepts like “safety” and promoting fear about new technologies, rather than taking the “purest view of law reform, which is to say what are the fundamental problems? How do we address those fundamental problems?”

Notwithstanding pragmatic efforts to align incentives, there will sometimes be irreconcilable misalignments between the immediate interests of a litigant and the public interest more generally. Bennet Moses put the problem in these terms:

Participants were quick to point out that many public interest clients are deeply committed to broader social justice outcomes. Nonetheless, a client’s interest in compensation and a speedy resolution of their complaint may be at odds with public interests in seeing a case through to judgment. In the worst case, Terry Carney pointed out, corporations and governments adopt a cynical strategy of separately settling individual cases – either in administrative review or in court – to obfuscate systemic problems and avoid judgments that produce change.

Transparency throughout the accountability lifecycle

The opacity of AI and ADM systems is well known. Scholars have pointed out different kinds of opacity – how algorithmic systems may be inscrutable, intentionally concealed, and difficult for those without technical expertise to understand (Burrell Reference Burrell2016; Henry et al. Reference Henry, Simcock and Snoswell2022; Selbst and Barocas Reference Selbst and Barocas2018). Our interviews added more color to this picture. They revealed not only that there are different kinds of opacity that obstruct accountability with respect to ADM – but that there are different moments along a timeline of accountability where opacity (or transparency) matters in different ways. We gained this insight, which is not emphasized in existing literature, as a result of our participant selection. Stakeholders from civil society, academia, and legal practice each contribute different stages of accountability-seeking with respect to ADM harms, and so could tell us about the role that transparency and opacity played in their own niche.

In the broadest sense, and throughout the whole lifecycle, opacity of and about ADM reinforces asymmetries of power. Subjects of unjust ADM may feel this viscerally. Lauren Perry, who had been undertaking qualitative sociological research into facial recognition technologies, and who was personally impacted by Robodebt, reflected that “for a lot of individuals, there’s this real sense that new and emerging technologies are happening to them.” In other words, opacity creates a sense of powerlessness.

Transparency about ADM is thus a pre-condition for accountability. Individuals, advocates, regulators, and the general public all need various kinds of transparency at multiple points along the accountability timeline to perform their various accountability functions. Participants were generally in favor of horizontal AI regulation mandating, at a minimum, transparency requirements for ADM. Interviews took place before the minor tranche of reforms to the Privacy Act in late 2024, and also before the Department of Industry, Science and Resources proposed (and then abandoned) the introduction of “mandatory guardrails” for high-risk AI, which would have imposed basic documentation, transparency, testing and risk management requirements (Safe and Responsible AI in Australia: Proposals Paper for Introducing Mandatory Guardrails for AI in High-Risk Settings 2024). As Kimberlee Weatherall has put it: this kind of regulation would create a “paper trail” that could then be used to enforce other laws (Weatherall Reference Weatherall2025).

Aside from direct explanations about ADM to users, participants reminded us that various kinds and degrees of transparency could provide a foundation for public interest legal action. Kate Bower, working at the time of our interview at CHOICE, a consumer advocacy organization, explained the value of transparency in facilitating accountability through “peer review”:

Without this kind of transparency about ADM deployment and function, it may be difficult or impossible even to detect ADM harm or understand its extent. Sam Floreani took the view that transparency measures such as documentation of ADM design and inspectability of algorithms would “enable [civil society] organisations like Digital Rights Watch to have an increased ability to see what’s going on and then respond or help people raise awareness.” As a discrimination scholar pointed out, this kind of transparency – mediated by journalists, academics, or public interest organizations – has the capacity to “raise community awareness about where these algorithms might be used, so they know where they might be impacted.” In short, transparency can operate on many levels. It can be directed at multiple audiences who might all play different roles in surfacing problems and progressing them toward appropriate accountability mechanisms, including litigation.

Each layer of transparency then facilitates further transparency and accountability mechanisms. Once there is sufficient transparency to identify a pattern of wrongdoing, to bring it to a lawyer, evaluate prospects of success, and bring a claim (all formidable challenges), litigation itself can shed further light on ADM harms. This was Vicki Sentas’ view: “I think its primary modest benefit is the exposure of data, which illustrates the harm.” On this view, litigation is itself a mechanism for transparency through its process of discovery and evidence gathering – and it is this role in bringing to light injustice which permits it to work effectively as part of an overall ecosystem of accountability and advocacy.

For litigation to work, however, lawyers need to understand wrongs and explain them satisfactorily to courts. Technical capability in the profession is therefore a condition for effective litigation. As Ed Santow put it, “there’s really important upskilling that needs to happen throughout the legal profession.” But Lizzie O’Shea reminded us that judges, despite generally coming from an older generation that is perceived to be less tech-savvy, are already heavily engaged with technology:

Min Guo suggested that enhanced technical capability in the profession might be a question of time, exposure, and familiarity. Since our interviews, the use of AI-based technologies has certainly become increasingly prevalent throughout the profession (Bell Reference Bell2025).

Technical capabilities around ADM could further be strengthened by involving expert witnesses. One human rights lawyer reflected, “there is good technical expertise out there … I’m sure you’d get expert evidence.” Lizzie O’Shea further explained,

To facilitate such upskilling, and to support a robust transparency-accountability lifecycle, the aggregation and sharing of information among stakeholders is essential.

Networks for information sharing and community mobilization

Transparency at various points in the accountability lifecycle can only lead to meaningful change if stakeholders are able to join the dots. Currently, the ecosystem for recognizing harm and progressing complaints is what one human rights lawyer called “fragmented,” with civil society organizations and regulators working in what consumer advocate Chandni Gupta called “silos.” Terry Carney talked about how systemic issues may slip through the gaps:

Participants had a range of suggestions about how to build what we came to think of as information aggregating mechanisms. One common thread was the urgent need for effective mechanisms for receiving, aggregating, and analyzing complaints to detect patterns of systemic wrongdoing. David Lindsay put it this way: “Litigation is the tip of the iceberg. It’s all of the complaints that people have day after day of navigating systems in which they feel disempowered.” A social justice litigator elaborated on how accessible complaint-making is a key condition for ADM accountability:

Lily Ryan, an information security specialist at Thoughtworks, noted that this could prompt further investigations through mechanisms such as freedom of information requests or inspections of code and algorithms.

Another mechanism for aggregating and distributing information is through appropriate allocation of regulatory responsibility. Terry Carney spoke of the need for institutions capable of systematically reviewing administrative complaints and processes, and argued for the reinstatement of Australia’s Administrative Review Council, which was composed of experienced decision-makers “much more attuned” to identifying administrative issues at a systemic level. Since the time of interviews, that body has in fact been reconstituted (Administrative Review Tribunal Act 2024).

Formal institutions with responsibility for systematic monitoring and review could therefore complement public interest litigation. Such institutions could modulate the problem that individual proceedings (whether for administrative review or judicial review) may be piecemeal and, on their own, inadequate to meet systemic harm. Regulators too have a role, as Chandni Gupta, Digital Policy Director at the Consumer Policy Research Centre, said:

Australia’s new National AI Plan established an AI Safety Institute to guide the development of safety practices and guardrails (National AI Plan 2025). While its remit seems at present to be highly technical – developing technical practices and guardrails to promote “safety” – there is an opportunity for the Institute to expand its impact by helping regulators in multiple domains develop better expertise and practices for identifying and responding to ADM harm at scale.

It is not enough, though, just to know about problems at a systemic level, or to rely on a review body or regulator to correct them. Part of the function of public interest litigation is as a “private regulatory tool” (O’Shea). No one knows the impact of digital harms better than those who suffer them. Community organizations have enormous expertise. Those who are affected by ADM also need to know how they can take action, and who will assist them. Part of this includes promoting legal literacy within communities. As Lauren Perry pointed out, “ways to increase community awareness around access to justice” is essential. This kind of network-building would also be an important path to address power dynamics whereby lawyers and those with specialist technical knowledge, rather than those more directly affected by injustice, come to dominate social movements (Handler et al. Reference Handler, Hollingsworth and Erlanger1978).

There is a foundational need for better solidarity networks and long-term relationship building, connecting would-be litigants, community organizations, and consumer bodies with the relevant resources and people. One human rights lawyer spoke about

Participants proposed various options to address this problem, including a specialist digital legal center and the development of effective referral networks. Ed Santow pointed out that the kinds of “hackathons” described above were an effective, informal way of developing these kinds of networks, from which collaboration, information sharing and referrals could develop organically: “it doesn’t need to have a lot of structure.”

Funding support for litigation and litigants

For public interest litigation to elicit transparency and accountability, it requires funding and support – as well as the removal of unnecessary cost barriers. Samantha Floreani and Jake Goldenfein both pointed out how relatively “immature” Australia’s civil society landscape was, compared to the US and the UK. Organizations that might aggregate information and coordinate action lack funding – which may be why our participants tended to see lawyers as key drivers of litigation.

At later stages in the accountability lifecycle, litigation cost support is also crucial. Community legal centers and Legal Aid need funding to smooth the transition from grievance to legal action. Min Guo pointed out that Legal Aid would require “endless resources,” and made the more radical proposal of developing a universal public legal services system, like Medicare. Other participants emphasized the need to ensure that private litigation funding remains viable.

But support must take forms other than direct funding. Many participants pointed out how obstructive the risk of adverse cost orders is to a would-be litigant. One serious problem is the fact that lead plaintiffs in class actions are solely liable for adverse cost orders (whereas any proceeds of the case are shared with the whole class). In one human rights lawyer’s words, “adverse costs orders against your lead plaintiff can stymy litigation or deter people from litigating.” A more generous arrangement regarding cost orders would significantly change the landscape – a view held by many of the litigators in our study. In the Robodebt case, it was the plaintiff law firm, Gordon Legal, that indemnified applicants against adverse costs orders, as well as acting on a no-win-no-fee basis. But a more systematic solution is warranted. While indemnities may help, getting them may be time-consuming and difficult.

Public interest exceptions to such orders would be one such option. Another option would be to provide that in public interest litigation against the government, only the government should have to pay adverse costs. Ed Santow pointed out that other jurisdictions have mechanisms of this kind, such as the protective costs orders available in public interest litigation in the UK. These arrangements have not, despite their critics’ concerns, opened the floodgates (Lock QC and Mills Reference Lock QC and Mills2016). All of the other burdens of litigation discussed here provide a strong disincentive against frivolous litigation. The UK might therefore provide a template for reform to Australia’s adverse cost regime.

On that note, for public interest litigation to be truly effective, litigants would also need support in meeting non-financial costs. Lizzie O’Shea explained that to properly support litigants under the extreme strain of litigation would require “a whole different social infrastructure and organisations to assist and work with them.” It makes sense to think of funding and support of accountability mechanisms in terms of ecosystems and lifecycles: not simply funding one action, but sustaining a whole infrastructure. A better-funded civil society could develop the structures, expertise, and systems to properly support litigants across the accountability lifecycle.