AI and the Coroners Court of Victoria: Concerns and Criticisms From Mob

Courts are slow in settler legal systems. The people in and around them move slowly and bear their emotional weight. The contested deaths that Coroners Courts examine often happen fast, and for First Nations deaths in custody they happen at escalating rates. AI doesn’t feel them like people do. Unlike the slowness of humans in courts, AI moves fast, also at an escalating rate.

AI was once a shadow cast over the future of court work, a cautionary tale from the United States that Australian jurists pointed to as a warning. It was contained to proprietary black-box decision-making even over there, handing out sentencing and parole decisions based on risk metrics that disproportionately incarcerated Black and Indigenous peoples and gave no easily-contestable basis. We do not have to look far to see the racialised impacts of AI decision-making (see for review, Baumer, 2013). Until recently and for the most part, the legal racism that First Nations people experience was organic. It came from those institutions, concepts and structures driven by people, or from colonial people and cultural and social systems themselves.

The use of AI in Australia and its courts has exploded in the last two years. No longer a cautionary tale, it is used to various effect by solicitors and barristers. A silk was reprimanded by Elliot J in 2025 for his use of Generative AI in a homicide matter involving a young person in the Victorian Supreme Court. The AI had hallucinated precedents that did not exist and the court could not find (ABC News 2025). Numerous courts have now issued practice notes and guidance on the use of AI (see, for example, Mortimer, 2026). We are writing this article because as of 2024, the Victorian Coroners Court itself is using AI, ostensibly for workload management and to provide distance from traumatic material.

We, the authors, have concerns around the use of generative^[1] and agentic^[2] AI in the especially sensitive coronial jurisdiction, a jurisdiction that has consistently fallen under the scrutiny of First Nations communities (Bennett, 2026; Hellmann, 2026). This article is a quick-fire summary of those concerns.

1. What is the coroners court?

Each state and territory across so-called Australia is its own coronial jurisdiction, and each has its own Coroners Court. In each jurisdiction, the coroner has legislated authority to investigate deaths that are sudden, unexpected, suspicious, or otherwise unnatural, as well as deaths occurring in specified circumstances such as accidents, suicides, homicides, and deaths in custody or state ‘care’. In performing this function, the coroner is required to ascertain the identity of the deceased and to determine the place, manner, and medical cause of death. The coronial jurisdiction is also oriented toward prevention, with coroners frequently expected to make recommendations aimed at reducing the likelihood of similar deaths occurring in the future (McCabe, 2024). Coroners Courts are settler legal institutions and, while First Nations Peoples have campaigned for their mandatory jurisdiction over reportable deaths (notably in the Royal Commission into Aboriginal Deaths in Custody), like all settler legal institutions they have serious constraints and often harms in the context of the colony (see e.g., Rule, 2019). We acknowledge those serious and troubling issues.

When a death occurs in custody there is a legislative requirement to hold a coronial inquest to examine the circumstances of the death; to uncover the ‘truth’ of what has happened (Scraton, 2013). First Nations Peoples are over-represented in every criminal jurisdiction on this continent, and die in custody at disproportionate rates (Australian Institute of Criminology, 2026). Each of these deaths necessitates a coronial inquest. Inquests are one of the very few avenues that First Nations families have to seek recognition of the harms caused by the State (McCabe, 2026).

1.1. What is the Coroners Court of Victoria using AI for?

In March 2024, the Coroners Court of Victoria (CCV) began implementing a bespoke AI program designed to reduce vicarious trauma and improve staff wellbeing (Coroners Court of Victoria, n.d.). The program was highlighted at the 2025 Asia Pacific Coroners Society Conference^[3] (the Conference). A demonstration of its capabilities was performed by one of the Amazon Web Services employees involved in the development of the program. Representatives from the CCV stressed the security of this program and the data it holds, and noted their adherence to various guides regarding the use of AI in legal practice. The potential for this program to reduce CCV staff exposure to distressing materials was offered as a key consideration in the development of the program, as well as its efficiency in ‘supporting coroners, solicitors and case investigators to analyse and review case material’ (Coroners Court of Victoria, n.d.). This AI model can remove or redact images; review large documents to identify key pieces of information; assist in preparing a legal brief; and may be used to capture and summarise sensitive conversations with bereaved families.

What was missing from the discussion at the Conference, and from discussions more broadly, is consideration of the specificity of the coronial jurisdiction, and the obligation of Coroners Courts to centre both the dead and the bereaved in all that they do. It remains unclear to us if those with lived experience on the other side of the coronial jurisdiction, or community leaders and Elders, were involved in the development of this program. Certainly, there was no public explanation of how this would relate to key concerns that First Nations people have about the Coroners Courts generally. These might include cultural protocols for AI use for First Nations deaths and post-death pathology, or how AI would work amidst the information and investigative imbalances that plague much of contested deaths of First Nations people (in custody or in healthcare). We are concerned that there has also been seemingly little attention given to the very real parallel threat of AI to Indigenous peoples: technological colonialism at large. What might this mean for death investigation and the coronial jurisdiction going forward?

2. Our concerns

2.1. Cultural flattening & technological colonialism

The failure of Generative AI to understand the nuances and intricacies of First Nations Peoples, communities, and experiences, is increasingly well-known. Adobe was identified as having a large database of AI-generated stock images of ‘Indigenous Australians’, images that included ‘clan markings’ with no cultural significance or grounding, and ‘Indigenous artworks’ with no cultural meaning or relation to any First Nations Person or community (Guenzler, 2025). Similarly, AI image generator Midjourney was found to produce images representative of the tired and offensive trope of the ‘noble savage’, amalgamations of First Nations Peoples from around the world, including Africa (McMullan & Stasiuk, 2025). This, according to McMullan and Stasiuk (2025), is a very specific form of cultural harm, one where AI fails to understand difference, resulting in more uniformity and less diversity – in other words, cultural flattening.

The term ‘technological colonialism’ is useful here. Put simply, it refers to the dominance of a small group of entities in specific geographic regions that, not necessarily purposefully, imbue AI programs with their own cultural values, biases, and norms (Dobrin, 2024). This in turn may well result in the homogenisation of cultural and social values globally. This technological colonialism is covert - ‘it is woven subtly into the fabric of AI systems. It manifests in how these systems interpret language, the cultural nuances they recognize or ignore, and how they navigate ethical dilemmas’ (Dobrin, 2024). A key concern here then is the inability of an AI program to understand the linguistic, cultural, and historically informed nuances not only present throughout police reports and witness statements, but in the conversations with those grieving their person (Perera et al., 2025).

AI is not, and cannot be, ‘ethical’ in its dealings with First Nations people. The technological colonialism inherent in its development will shape cultural narratives in ways we cannot fully anticipate, eroding cultural identities (Dobrin, 2024). This ‘risk of cultural homogenization’ is accelerated by the embedding of AI into every facet of our lives, and now, via the CCV, our deaths (Dobrin, 2024).

2.2. Indigenous Cultural and Intellectual Property

The use of Indigenous Cultural and Intellectual Property (ICIP) is governed by a specific set of protocols (Janke 2025). The CCV, like other Coroners Courts, sees, documents, and retains incredibly sensitive material. This includes evidence that may be sensitive in terms of public or gendered access, or that may be sensitive in terms of its discussion of death, remains and dying.

We have found no evidence that ICIP protocols were considered in the development of this bespoke AI program at the CCV. While there remains no data collection and AI training exception to copyright rules in Australia, copyright in court decisions and other legal instruments^[4] is not infringed by making ‘one copy’ ‘for a particular purpose’ (Copyright Act 1968 s182A). Matters before the Court, then, are vulnerable to data harvesting by for-purpose generative and agentic AI, despite the assurances from the CCV that the data accessed by the program is held on ‘a secure, isolated system with no private data passed to other entities’ (Coroners Court of Victoria, n.d.).

2.3. Indigenous data sovereignty

Indigenous data sovereignty is one branch of ICIP. In Indigenous data sovereignty, First Nations communities retain control over data relating to them. The Maiam nayri Wingara Indigenous data sovereignty collective describe control as ‘expressed through the creation, collection, access, analysis, interpretation, management, dissemination, and reuse of Indigenous Data’ (Maiam nayri Maiam Nayri Wingara, n.d.). Historically, universities and other public research institutions gathered large volumes of data about First Nations people which those communities rarely had access to and often had used against them (see Lovett et al., 2019). This includes administrative bodies and specialist courts and tribunals, like the Coroners Court.

Indigenous data sovereignty is, at least at the time of writing, incompatible with the use of generative and agentic AI. This is especially the case for sensitive data concerning death. There are many reasons for this, among them the inability of AI to consistently report back on its internal decision-making, resulting in a lack of not only community control but meaningful human control over the use of Indigenous data. Other concerns include the extraordinary trust required to believe that Amazon Web Services will not in some way derive training or insights from the data it is given, even in closed systems.

A recent report has also identified some of these issues: an Indigenous-led study examined the ways in which First Nations Peoples in so-called Australia are encountering and responding to AI (Carlson & Worrell, 2026). For the participants in this study, AI is ‘understood as part of broader systems of extraction, surveillance, and control’, ‘where errors or bias can have serious consequences’ (Carlson & Worrell, 2026, p. 8). Concerns about data sovereignty were identified as being ‘immediate and unresolved’, and ‘incompatible with Indigenous knowledge systems’ (Carlson & Worrell, 2026, p. 21).

However, adjudicative and administrative bodies can, with care and forethought, build Indigenous data sovereignty into how they do business. The Treaty Authority (2025) in Victoria, for instance, requires access to Indigenous data for particular communities if they want to advance in the Treaty process. Acknowledging their ownership of their data, they have developed a data sovereignty framework that helps communities trust and control how data is created, managed and applied. While not framed as data sovereignty, the Federal Court has developed practices around sensitive Indigenous knowledges and data. It is not outside the realm of possibility for the CCV to do so, too. There is no mention of sensitive Indigenous data and AI protocols around it in its Practice Direction (6 of 2020) on Indigenous Deaths in Custody, the CCV’s only Indigenous-specific practice direction.

2.4. Tensions in disputes and scope

The AI we were presented with at the Coroners Conference by Amazon Web Services was showcased as being able to read a brief of evidence and pull out particular points of interest for a coroner or their staffers to pay attention to. These points of interest were presented as contradictions in witness statements (for example, a car turning left versus right). These are not commonly matters that receive coronial attention at inquest, at least in terms of the matters First Nations people bring to those courts. Rather, even among elective inquests, the focus remains on matters less obvious and sometimes outside of the attention of witnesses. In fact, Victorian First Nations families have advanced issues at inquest precisely because they were outside the conscious attention or documentation of individuals and investigators: namely, systemic and structural racism. See, importantly, the advocacy of the Day family in the Inquest into the death of Tanya Louise Day (Coroners Court of Victoria, Coroner Caitlin English, 2020); and the advocacy of the Nelson, Walker and Lovett families in the Inquest into the passing of Veronica Nelson (Coroners Court of Victoria, Coroner Simon McGregor, 2023). We are alarmed that focussing on direct contradictions and similar, obvious points of interest in the brief, potentially undermines the more subtle and exploratory parts of the jurisdiction that are essential to its function and design, especially for mob.

2.5. ‘Black box’ systems that remain unaccountable

Most AI models can be described as ‘black box’ systems. The way in which AI operates is so complex that even those who created and trained the systems are not entirely sure how they work. AI models are commonly either unable to account for how they came to an answer or misreport how they came to an answer. While large language models can occasionally explain their behaviour, they are not always honest and have been known to hallucinate sources (Choi & Mei, 2025).

The legislated functions of coroners across Australian jurisdictions are primarily ‘fact finding’; if the coroner in any given matter could not explain how a decision or finding came to be, the core function of the court would be undermined. Why then is it appropriate to use AI in this jurisdiction, when its reasoning cannot be consistently accounted for? This is particularly troublesome when AI is used to summarise key points in a brief, or to synthesise evidence into a case narrative, at the early stages of an investigation or inquest when key decisions are being made about scope, witnesses and court resourcing. We understand that the AI is not itself making substitute decision-making for the coroner (and the CCV insists it ʻdoes not override judicial decision making or independence’ (Coroners Court of Victoria, 2025, p. 3)), but it is nudging and defining their attention at critical points. While it is not substituting judicial decision-making, it’s hard to see how the following functions, disclosed in a submission to the Victorian Law Reform Commission, wouldn’t influence Coroners’ core work:

• generating written descriptions of ʻdistressing images’

• summarising matters to inform news organisations about upcoming inquests

• preparing chronologies from ʻa coronial brief and other source materials’ (Coroners Court of Victoria, 2025, p. 3)

• extracting data from the coronial brief and identifying ʻevidentiary gaps’ and locating ʻrelevant evidence’. (Coroners Court of Victoria, 2025, p. 3)

2.6. Entrenching (data) patterns based on previous data

Generative and agentic AI are not autonomous, creative thinkers in the way that humans are. They are probability models trained on pre-existing data. There has been a recent and welcome tendency in Coroners Courts of both institutional and decision-making innovations. They include the development of Aboriginal support units within Coroners Courts, the co-housing of the Victorian Institute of Forensic Medicine to provide forensic research to the CCV directly, and the slow expansion of the scope of relevance at inquests, particularly for First Nations deaths. The use of AI, which by its nature can only repeat and cannot innovate, limits new opportunities for this growth and stymies new visions of the possible within Coroners Courts, including those driven by First Nations People as interested parties. This is especially the case when, we can only assume, the data it may draw on internally from the CCV are briefs of evidence prepared by state parties, whose dominance of inquests into contested Indigenous deaths has been the subject of First Nations public scrutiny for decades.

2.7. The Right to Participate

Families forced to engage in the coronial jurisdiction do so on the basis that their participation is prescribed and restricted by the rules of the coronial process. Families of the deceased often have little to no influence on how the death of their loved one is investigated, the scope of an inquest, what story gets told of their loved one by state agencies, or what findings are made by the coroner. Further, the expectations imposed by court etiquette standards mean that families must control how they experience their grief and loss, in real time, or risk being further restricted in their participation and, in some cases, denied their right to participate in the courtroom through punitive measures such as being asked to leave the courtroom, and being prevented from reading family impact statements (Morse, 2023). All this to say, families of the deceased are already forced to participate in a process that is deeply limited in its capacity to be human, and hold grief and loss, despite being a process intimately connected to death.

We are concerned that the introduction of an AI program to further reduce human engagement in the coronial process, through limiting exposure to sensitive materials and optimising the coronial process, risks further undermining how families of the deceased can participate in the coronial process. The use of AI technology to analyse and synthesise evidence at various stages of the inquest process implies that the technology can have a significant influence on how the matter is investigated, and what is submitted to and ultimately considered by a coroner. As discussed earlier in this article, there is concern that AI generated summaries or reports cannot account for cultural nuance, history and the ongoing impact of systemic racism within institutions, furthering flattening the level of inquiry required by the court, particular in matters brought to the attention of the court by First Nations people (Perera et al., 2025). This flattening will further limit the extent in which issues important to families of the deceased are considered by the court. Reducing opportunity for the critical, and what can be complex, human inquiry that is arguably required of, and central to the integrity of the coronial process, furthers the existing constraints within the jurisdiction, while also increasing the depth and range of harm that may be experienced by families of the deceased throughout the coronial process. This is particularly concerning for families in which the inquest process is their only opportunity to try to understand how their loved one passed, and/or pursue institutional accountability and/or redress on behalf of their loved one, albeit limited.

For First Nations people the implementation of AI technology without seeking consent to when and how it is being used explicitly undermines First Nations’ rights to free, prior and informed consent, to participate in decision-making in matters which would affect their rights, and our entitlement to just and fair redress as affirmed in the United Nations Declaration on the Rights of Indigenous Peoples (UNDRIP 2007).

2.8. Environmental impacts on Country

While the conference did not speak on the environmental impact of implementing the AI pilot into the CCV, it is important that any discussions around AI technology include consideration of their impact or potential impact on Country. AI programs rely on resource intensive physical and digital infrastructure to operate. In particular they require significant demand on water and energy as power sources. As raised by Bundjalung woman and CEO of Current Country Social Change, Arabella Douglas, ‘water is not a neutral input. It is living Country’ (Currie Country Social Change Aboriginal Corporation, 2026, p. 3[4]). First Nations led submissions to the NSW Parliamentary Inquiry into data centres in NSW have characterised data centres as ‘resource-intensive industrial developments’ that present ‘cumulative risk’ for water, energy systems, Country and community and called for the implementation of First Nations governance and First Nations’ decision-making in the expansion of AI and its required infrastructure (Currie Country Social Change Aboriginal Corporation, 2026; Tranby Aboriginal Co-operative Ltd., 2026). Further, submissions have recognised that the expansion of data centres across the continent without First Nations leadership perpetuates ‘structural and rights dispossession’ of First Nations peoples and Country (Currie Country Social Change Aboriginal Corporation, 2026). The development and implementation of AI programs cannot be silent on the extractive relationship these technologies have on Country and should include First Nations communities as decision-makers from the outset.

3. Conclusion

The introduction of AI into the Coroners Court of Victoria represents a profound shift in how death, evidence, and grief are mediated within a jurisdiction that is uniquely charged with public accountability, truth-finding, and care for the bereaved. Coroners Courts pride themselves as frontiers of therapeutic jurisprudence. Core to therapeutic jurisprudence is the dignity of the harmed and bereaved who appear before the Court (Perlin, 2019). AI cannot adequately hear the concerns of families and communities around contested death, nor can it honour a story nor offer empathy and express grief. The digital bureaucratisation of death is deeply concerning for human dignity. While the stated aims of the CCV AI pilot, namely, reducing vicarious trauma and improving efficiency, are not insignificant, they cannot be assessed in isolation from the social, cultural, legal, and ethical obligations that structure coronial work. In a jurisdiction that disproportionately examines the deaths of First Nations Peoples, and which has been the subject of justified scrutiny by First Nations communities, this uncritical adoption of AI risks reproducing and entrenching the very harms coronial inquests/investigations are meant to address.

AI systems are not neutral tools. They are culturally situated, probabilistic technologies that reflect the values, assumptions, and power relations of their designers and training data. In the coronial context, this manifests as risks of cultural flattening, technological colonialism, erosion of ICIP, diminished human dignity, and the displacement of relational practices that are central to the jurisdiction’s purpose. The opacity of ‘black box’ systems sits in direct tension with the coronial mandate to explain how findings are reached, while the reliance on historical data threatens to calcify existing patterns rather than enable the critical innovations that coronial reform has increasingly sought.

Particularly concerning is the apparent absence of free, prior and informed consent, and the lack of First Nations governance in the development and implementation of the CCV AI pilot. The use of AI to analyse evidence, synthesise sensitive materials, and potentially assist in judicial decision-making constitutes a material intervention in the coronial process. Such interventions risk further marginalising those who already experience the coronial jurisdiction as alienating, and extractive. These risks are compounded by digital exclusion and by the environmental impacts of AI infrastructure on Country, raising questions not only about justice for those who have died, and the bereaved left behind, but about justice for land, water, and future generations.

AI programs may have a role to play in supporting coronial staff, but its use cannot, and should not attempt to, substitute or override human judgement. Nor can it be introduced without rigorous scrutiny of whose knowledge is valued, and who bears the risks of technological failure. If coronial jurisdictions are to engage with AI at all, they must do so cautiously, transparently, and under First Nations leadership, grounded in Indigenous data sovereignty principles, ICIP protocols, and UNDRIP obligations. Anything less risks transforming the coronial process into yet another site where technological efficiency is privileged over truth, care, and self-determination, a site where the bureaucratisation of death comes at the cost of human dignity.