33 Indigenous Australians died in custody in 2024–25, the most on record — a study of 134 deaths since 1991 found 5 referrals to a prosecutor and no convictions

What is happening?

Thirty-three Aboriginal and Torres Strait Islander people died in custody in Australia in 2024–25, the highest annual figure since national counting began in 1979–80. Twenty-six of those deaths occurred in prison custody, up from 18 the year before.

Australia counts these deaths and examines them. A national deaths in custody program has run since 1980 and has been operated by the Australian Institute of Criminology since 1992, and every death in custody or during police contact draws a coronial inquest by law. There is no gap in the measuring stage.

What is missing sits after the inquest. A study published in 2018 reviewed 134 Aboriginal and Torres Strait Islander deaths in custody that occurred after the 1991 Royal Commission. In 11 of those cases a coroner considered referring the matter to a prosecutor. In 5 the referral was made. Two reached a courtroom, and both ended with the charge withdrawn or with an acquittal, so the sample produced no conviction. Those counts describe that sample as stated in June 2020, and no source opened here updates them for deaths since.

The cumulative toll is recorded in two snapshots that do not sit on the same date. Reporting in December 2025 cited at least 618 Aboriginal and Torres Strait Islander deaths in custody since the Royal Commission concluded in 1991. A separate source citing the same national series in June 2020 gave 432 since 1991. Whether the two rest on the same counting basis is not stated by anything opened here, so the difference between them is not treated as a number of deaths.

The remedy proposed in 1991 has been scored rather than finished. The Royal Commission into Aboriginal Deaths in Custody made 339 recommendations across 26 themes and 547 sub-items. A review commissioned by the federal Minister for Indigenous Affairs in 2017 and delivered by an external consultancy in 2018 assessed 78 percent as fully or mostly implemented, 16 percent as partially implemented and 6 percent as not implemented. In 2021 five researchers at an Australian university argued that the scope and method of that review overstated implementation, and 33 scholars and experts signed the challenge. Reporting in December 2025 described implementation as standing below two thirds. All three assessments are still in circulation.

The population exposed to all of this is growing. On 2025-06-30 the age-standardised Aboriginal and Torres Strait Islander adult imprisonment rate was 2,500.2 per 100,000, against 2,318.4 a year earlier and a 2019 baseline of 1,925.4 that a national target undertakes to cut by at least 15 percent by 2031. The Productivity Commission assessed that target as worsening, at high confidence.

Whose problem is this?

The body that holds the inquest is not the body that could bring a charge, and neither of them is the body that answers for whether the recommendations of 1991 were carried out. Each of the three can complete its own task while the sequence as a whole produces nothing.

Where does this problem end?

One boundary is genuinely unsettled and it matters for the count. A death during a restraint on a supermarket floor is not obviously the same kind of event as a death in a cell, and no source opened here states how the national series classifies it.

A second boundary could not be drawn at all. Whether a death is recorded as arising from an existing illness or from the immediate use of force is a division that none of the sources opened here supplies consistently, so this document does not describe the causes of the deaths it counts.

What is the state now, and what should it be?

Now

Australia measures this problem closely and publishes the counts on a fixed cycle. What the measuring has not produced, across the one sample that has been studied, is a single conviction.

Needs a new measurementthe target state: one official target exists and it describes the denominator rather than this problem. It undertakes a reduction of at least 15 percent in the adult imprisonment rate by 2031 against the 2019 baseline of 1,925.4 per 100,000, and the rate measured on 2025-06-30 was 2,500.2, roughly 30 percent above that baseline rather than below it. No source opened here states a target for the number of deaths in custody, a target for how many coronial findings should become charges, or a date by which the 339 recommendations are to be complete. The recommendations themselves describe the intended floor, and every source opened here is silent on when that floor becomes binding on anyone.

How big is it?

17,432 Aboriginal and Torres Strait Islander adults were in prison in Australia on 2025-06-30. That is 37 percent of the 46,998 adult prisoners counted that day, in a population that stood at 983,700 people or 3.8 percent of the country as of June 2021 and that passed one million on an estimate published in 2024. The count rose by 1,561 people, or about 10 percent, in the year to that date.

That figure is a count on one day and it is the only national count of people in Australian custody found in the sources opened here. It leaves out three groups. It leaves out police custody entirely. It leaves out youth detention. And it leaves out everyone who entered and left custody at some other point in the year, which a single-day count excludes by construction.

Only a fragment of the first omission can be measured. About 16,000 Aboriginal and Torres Strait Islander people a year passed through the custody notification service in New South Wales and the Australian Capital Territory on an estimate made in 2017. That covers two of eight jurisdictions and it is an annual flow rather than a count on a day, so it cannot be added to the prison figure without producing a number that corresponds to nothing.

The second omission cannot be measured at all from these sources. Aboriginal and Torres Strait Islander 10 to 17 year olds were held in detention at an average daily rate of 25.7 per 10,000 in 2024–25, and the population that rate applies to is not given anywhere here, so the number of young people behind it cannot be derived.

Exposure and harm are different quantities and both belong here. The exposure is about 17,000 adults on any given day and rising. The harm was 33 deaths in 2024–25 and at least 618 cumulatively as cited in December 2025, figures smaller by three to four orders of magnitude that answer a different question.

Under what conditions does it arise?

1. The finding stage is compulsory and the consequence stage is discretionary. A coroner establishes how a person died and may make findings, recommendations and referrals. Only a prosecutor can lay a charge, and only where the evidence would establish a criminal offence to the criminal standard. Nothing in the sequence obliges the second body to act on the work of the first, and in the sample of 134 deaths that has been studied the sequence produced 11 considerations, 5 referrals and 2 court matters.

2. A structural finding can arrive after the criminal question has already closed. In the shooting death of an Aboriginal man in the Northern Territory in 2019, the officer was acquitted at trial in 2022 and dismissed from the force in 2023. On 2025-07-07 the coroner found that the officer held racist views, that there was institutional racism in the local police station, and that the death was avoidable. The finding that named the structure came three years after the last point at which it could have borne on a charge.

3. Recommendations without a deadline become a score that can be argued about indefinitely. Governments accepted 339 recommendations in 1991. Thirty years later one commissioned review put implementation at 78 percent, academic reviewers said the scope and method overstated it, and later reporting put it below two thirds. No body is charged with settling the difference, and a disputed percentage is easier to live with than a missed date.

4. The one measure with a visible effect on deaths was adopted one jurisdiction at a time. A custody notification service was legislated in New South Wales and the Australian Capital Territory in 2000, and no Aboriginal death in police custody was recorded in those two jurisdictions across the following 16 years. The death that occurred in 2016 followed a failure by police to make the notification. Queensland refused the service in 2017, and the Northern Territory adopted a version that excluded protective custody and arrests made without paperwork.

Each of these four conditions describes a handoff that nobody owns. None of them describes a fact that is unknown.

What has been tried?

The attempt with the clearest measured effect on deaths was never extended to every jurisdiction, and the two national efforts produced a list of recommendations and a numerical target rather than a duty that anyone can be held to.

What was found?

Why is it still unsolved?

Enforcement absent — the stage that establishes what happened is compulsory, the stage that could impose a consequence is discretionary, and nothing joins the two.

This is not a counting failure, and the distinction is the whole point of the document. A national program has recorded deaths in custody since 1980, an inquest is mandatory for every death in custody or during police contact, and the annual figures are published. Australia knows how many people died and it knows a great deal about how. The record then stops moving. In the one sample that has been examined, 134 deaths produced 11 considerations of a referral, 5 referrals, 2 court matters and no conviction. Knowledge is not the missing input.

The second reason is that the rate at which findings become charges is not a measured quantity. A prosecutor is under no obligation to explain a decision not to charge and no source opened here reports how often that decision is taken, which is why the only figures available come from an academic sample of 134 deaths stated as of June 2020 rather than from an official annual series. A quantity nobody publishes cannot become a target, and a stage nobody measures cannot be shown to be failing.

The third reason is that the remedy of 1991 was written as advice. The 339 recommendations were accepted by governments that were given no deadline and no auditor with the power to close the list. A review commissioned by the same governments scored implementation at 78 percent in 2018, academic reviewers said that overstated it, and later reporting put it below two thirds. Three answers coexist because nothing forces a fourth, and each new anniversary produces another assessment instead of another consequence.

The fourth reason keeps the pattern stable. The denominator is moving the wrong way at the same time. The adult imprisonment rate reached 2,500.2 per 100,000 on 2025-06-30 against a baseline of 1,925.4 set in 2019 and a target 15 percent below it by 2031. More people held, a mandatory inquest for every death and a discretionary route to any charge together produce more findings and no more consequences, year after year, with nothing in the process registering that combination as a failure.

What observation would mean it is solved?

Candidates — (a) an official national series is published showing how many deaths in custody are referred to a prosecutor and how many referrals become charges (b) the Aboriginal and Torres Strait Islander adult imprisonment rate falls below the 2019 baseline and keeps falling (c) a custody notification duty operates in every jurisdiction without carve-outs and the annual count of Aboriginal and Torres Strait Islander deaths in custody falls across several consecutive years.

(a) measures process rather than outcome, and it is still the one to build first. A published referral rate would end the position where the only figures available come from one academic sample of 134 deaths stated as of June 2020, and it would make the question answerable every year instead of once a generation. It would not by itself change any outcome, and a rate can be published and sit at zero.

(b) is the denominator. A falling imprisonment rate reduces the number of people exposed, which matters on its own terms and is what the national target measures. It says nothing about what happens after a death, and it has moved in the opposite direction in every year since the baseline was set.

(c) counts deaths, and the count moves for reasons that have nothing to do with accountability. In a country holding about 17,000 Aboriginal and Torres Strait Islander adults in prison, an annual total in the low tens can swing on a single year. The 16-year record in the two jurisdictions that legislated first is the most suggestive evidence in this document that a specific duty changes a specific outcome, and it rests on an encyclopedia entry rather than on an evaluation, which is why it is graded low above. It also marks the limit of that measure: it reaches the police custody stage, and 26 of the 33 deaths in 2024–25 occurred in prison custody. The three have to be read together, and (a) is the one that would let anyone check the other two.

What is it connected to?

Fills with researchbail and sentencing law in each state and territory, which sets the size of the custody population that this problem is measured against; the funding and coverage of Aboriginal and Torres Strait Islander legal services; health care inside custodial settings, which bears on deaths recorded as arising from illness; police disciplinary processes as distinct from criminal ones; and the same sequence in other countries where an Indigenous population is held at a disproportionate rate. Relation type and evidence grade were not confirmed in this round.

What these sources do not say

• What the national statistical tables actually contain. The figures of 33 and 26 for 2024–25 come from reporting that cites the national deaths in custody program. Every automated attempt to open that program on its own domain timed out during research, so no breakdown by custody type or by jurisdiction was read here. This is a failure of access rather than an absence of data, and it is recorded as such in the evidence table below.
• Whether 618 and 432 count the same things. The two cumulative totals are snapshots of the same national series taken at different dates, and no source opened here states whether they share a scope on youth detention or on how the territories are counted. The gap between them is not treated in this document as a number of deaths in an interval.
• Whether the referral figures still hold. The counts of 11 considered, 5 referred, 2 reaching court and none producing a conviction come from a 2018 study of 134 deaths and were stated as of June 2020. Nothing opened here updates them for deaths since, so they describe that sample at that date and cannot support a claim about the position today.
• Where the two implementation assessments diverge. One review scored 78 percent and later reporting put the figure below two thirds. The scoring method that produces the difference was read only in summary here, because the paper challenging the review was not opened in full.
• What Queensland and Tasmania do now. The position confirmed here on a custody notification service in those two jurisdictions dates from 2017 to 2020. No source opened here states the position in 2026.
• How deaths divide between illness and the immediate use of force. No source opened here gives that split at a national level, so this document does not describe the causes of the 33 deaths counted in 2024–25.
• What happened in a death recorded in May 2025. An Aboriginal man died after being restrained by plainclothes police officers in a supermarket in Alice Springs on 2025-05-27. As of the December 2025 reporting opened here, more than six months later, no official account had been settled, including whether the event counts as a death in custody at all.

See the evidence

No primary Australian statistical series was read directly. The two figures that carry the headline, 33 deaths in 2024–25 and 26 of them in prison custody, come from reporting that cites the national deaths in custody program, and every automated attempt to reach that program timed out, so the row for it in the table above is blank with the reason given. The government dashboard and the Bureau of Statistics release were opened directly and carry the imprisonment rate, the baseline, the assessment of worsening, the count of 17,432 and the youth detention rate. The government agency page for the implementation review carries the 339 recommendations and the 78 percent assessment. The referral and conviction counts rest on a single opened source describing a study published in 2018, and they are stated there as of June 2020; that scoping is kept in every place those numbers appear in this document, because the headline of the source itself makes a wider claim than the study can carry. Where sources overlap they agree in direction: the imprisonment rate appears in two opened sources with the same value, and the position of the custody notification service in each jurisdiction appears in two. Where they disagree the disagreement is left visible rather than resolved, which covers the two cumulative totals of 618 and 432 taken at different dates with counting bases that were never compared, and the two assessments of implementation at 78 percent and below two thirds. One of the sources in the table is an encyclopedia entry and it is graded low wherever it is the only support for a claim. This is a Path A output, so observation_refs is empty and provenance_mode: press-derived.

People affected

Estimated range 17,432–17,432 As of 2025-06-30

Sensitivity The interval has zero width and is not a confidence interval. It is a single published count on one day, and no source opened here restates it, projects it forward or attaches a margin to it. The correction runs in one direction only, upward, and it could not be computed from the sources opened. Three groups sit outside the count. Police custody is not counted here at all: the only measure found covers two of the eight Australian jurisdictions, being about 16,000 Aboriginal and Torres Strait Islander people a year passing through the custody notification service in New South Wales and the Australian Capital Territory as estimated in 2017, and that figure is an annual flow rather than a count on a day, so adding it to a single-day prison count would produce a quantity that corresponds to nothing. Youth detention is not counted here either: Aboriginal and Torres Strait Islander 10 to 17 year olds were held at an average daily rate of 25.7 per 10,000 in 2024–25, and no source opened here gives the population that rate applies to, so the number of young people behind it cannot be derived. Everyone who entered and left custody at some other point in the year is excluded by construction from a single-day count. The count is also rising rather than stable, having grown by 1,561 people or about 10 percent in the year to 30 June 2025, so it will be low for any later date. Against all of that the count measures exposure and not harm. The harm counted separately was 33 deaths in custody in 2024–25, the highest annual figure on record, and a cumulative figure of at least 618 since 1991 cited in December 2025, quantities smaller by three to four orders of magnitude that answer a different question.

Regional breakdown No source opened here gives a state and territory split of the count of Aboriginal and Torres Strait Islander adults in prison. Splitting the national figure by population would be proportional allocation, and it would be wrong by a wide margin in both directions at once, because imprisonment rates differ enormously between Australian jurisdictions while the Aboriginal and Torres Strait Islander share of the resident population differs on a different pattern again. The only jurisdiction-level facts confirmed in this round are policy positions rather than counts, being that a custody notification service operates in six jurisdictions, that the version adopted in the Northern Territory excludes protective custody and arrests made without paperwork, and that Queensland refused the service in 2017.