Three separate investigations into the Robodebt scheme produced 97 code of conduct breach findings and two findings of serious corrupt conduct — by March 2026 none of them had produced a referral to prosecutors and no criminal charge had been laid

What is happening?

Between 2015 and 2019 the Australian Commonwealth ran an automated welfare compliance programme, known publicly as Robodebt, that averaged annual income data to raise and pursue debts against welfare recipients. About 450,000 people were issued a debt under it. A Royal Commission examined the programme and delivered its final report on 2023-07-07 — three volumes, 990 pages, 57 recommendations, and a separate 56-page sealed chapter.

The sealed chapter did what the rest of the report could not. It referred six individuals to the National Anti-Corruption Commission: five current or former Commonwealth officials, called here Officials A to E, and a former senior minister. It made two further referrals that are frequently conflated with that set and are not part of it — Official C and another official who is not among the six were referred to the ACT law society, and Official E was referred separately to the Australian Federal Police over suspected perjury.

Every one of those tracks has now closed, and the record of what followed is as follows.

On 2024-06-06 the anti-corruption commission decided not to investigate any of the six, on the basis that the Royal Commission had already examined the conduct sufficiently. On 2024-07-24 the federal police closed the perjury investigation, stating that the evidence did not sufficiently establish an intention to mislead the Royal Commission. On 2024-09-13 the Australian Public Service Commission reported on the code of conduct: 12 current and former officials were found to have breached the code a total of 97 times, with 25 of those breaches attributed to two former departmental secretaries, one with 12 and one with 13. Both had already left the service when the finding was made, and no disciplinary sanction was imposed on either.

On 2024-10-30 the Inspector of the anti-corruption commission found that the head of that commission had declared a conflict of interest in relation to one of the six and had then failed to withdraw properly from the decision. The Inspector characterised that as officer misconduct and recommended the decision be reconsidered. A former High Court judge was appointed in December 2024 to conduct that reconsideration, and on 2025-02-10 decided the six referrals should be investigated after all.

On 2026-03-11 the anti-corruption commission published the outcome of that investigation, conducted under the name Operation Myrtleford. Two of the six were found to have engaged in serious corrupt conduct, over the preparation of a 2015 submission to cabinet and over the concealment of legal advice from an ombudsman inquiry in 2017. The other four were cleared, including the former senior minister. Neither of the two was referred to prosecutors: the commission assessed the evidence as insufficient to prove the conduct beyond reasonable doubt. It made no further recommendations of its own, on the view that the recommendations already made by the Royal Commission were sufficient to prevent a repeat.

The civil track moved on a different schedule and produced money. On 2026-06-23 the Federal Court approved a settlement covering about 125,000 registered claimants, comprising 475 million Australian dollars in compensation plus up to 13.5 million in legal costs, reported as a package of 548.5 million. Total Commonwealth spending attributed to the scheme exceeds 2.4 billion Australian dollars.

Whose problem is this?

The bodies that decide what follows a finding are not the bodies that made the finding, and no single institution in this sequence both establishes what happened and determines what happens next.

Where does this problem end?

The boundary here separates a finding of fact from a consequence attached to that finding. Every body in this record produced the first and none of them produced the second, and this document measures the distance between those two things rather than the quality of any single investigation.

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

Now

Thirty-two months separate the referral of the six from the publication of the outcome, and across that interval the only measure that moved in the direction of a consequence was the compensation paid by the Commonwealth to the people the scheme had pursued.

Needs a new measurementthe target state: no source opened here states what outcome the referral process was meant to produce, what proportion of corruption findings is expected to reach a prosecutor, or what sanction is available against a person who has left the public service. The Royal Commission set no target when it made the referrals, the anti-corruption commission published no standard against which its own decision not to refer could be measured, and the public service commission described what it found without saying what it thought should follow. Every source describes the sequence of decisions and none of them names a benchmark that the sequence could have failed.

How big is it?

About 450,000 people were issued a debt under the scheme, and that is the population the accountability process was convened to answer for. The figure is an estimate carried by a source opened in research rather than a count published by a primary body, and the research also recorded other coverage using values from 433,000 to 526,000 without any of those being traced to a source that could be opened. The population layer records the estimate with that disagreement stated beside it.

The accountability quantities are several orders of magnitude smaller and answer a different question. Six individuals were referred for corruption investigation, 12 were found to have breached the code of conduct 97 times, two were found to have engaged in serious corrupt conduct, one criminal investigation was opened and closed, no referral to prosecutors was made, and no criminal charge has been laid. The disciplinary count that would sit beside those is partly unavailable: two of the 12 were beyond sanction because they had already left the service, and no opened source states what happened to the other ten.

The money moved on its own scale. About 125,000 people registered as claimants in the settlement approved in June 2026, which carries 475 million Australian dollars in compensation and up to 13.5 million in legal costs and is reported as a package of 548.5 million. Those two components add to 488.5 million rather than to 548.5 million, and no source opened here reconciles the difference. Total Commonwealth spending attributed to the scheme is reported at more than 2.4 billion Australian dollars.

Reading the two sets together is the point of this document. Roughly 450,000 people were pursued for debts under an unlawful programme, and the entire consequence borne by any individual responsible for that programme is the publication of a finding.

Under what conditions does it arise?

1. The standard of proof rises as the consequence gets heavier, and the heaviest one sits at the end. A Royal Commission establishes what happened. A code of conduct investigation finds a breach on an administrative standard. An anti-corruption commission makes a finding of corrupt conduct on its own statutory test. A criminal charge requires proof beyond reasonable doubt. Findings accumulate at every level below the last one and then stop at its threshold, which is exactly what the assessment published in March 2026 records — conduct found to be seriously corrupt, and evidence assessed as insufficient to prove it to a criminal standard.

2. Discipline reaches employees, and the most senior people had stopped being employees. The code of conduct binds people in the service. Both of the former departmental secretaries who accounted for 25 of the 97 breaches had left before the finding was made, so no sanction was available regardless of what the investigation concluded. Availability of the disciplinary instrument therefore depends on the timing of a departure rather than on the seriousness of the conduct, and an investigation that takes years will routinely arrive after the people at the top have gone.

3. The body that decides whether to refer a finding for prosecution is the body that made the finding. No source opened here identifies any external review of the decision not to refer the two adverse findings to prosecutors. The assessment of evidentiary sufficiency and the interest in how the investigation is judged sit inside the same institution.

4. Overlapping bodies can each treat another as having dealt with it. The first refusal to investigate rested on the reasoning that the Royal Commission had already covered the conduct. That reasoning substitutes an inquiry which cannot impose a consequence for a body which can, and it took an inspector finding, an independent reconsideration and eighteen months to displace.

5. The oversight of the oversight body produced the same shape of outcome. The Inspector found officer misconduct in the handling of a declared conflict of interest by the head of the anti-corruption commission. What followed was a reconsideration of the underlying decision, and no source opened here reports any sanction. The mechanism built to correct the corruption body reproduced the pattern the corruption body was being corrected for.

6. The compensation track and the accountability track never meet. Money moved through the Federal Court on its own timetable and reached about 125,000 registered claimants. Nothing in that process determines whether anyone is responsible, and nothing in the accountability process is affected by how much has been paid. A government can therefore be seen to have answered for a programme in full while no person has answered for it at all.

What has been tried?

What was found?

Why is it still unsolved?

Enforcement absent — the institutions exist, they investigated, they found against people, and nothing in the design of any of them turns a finding into a consequence.

This is not a case of a missing body. Australia built a federal anti-corruption commission, gave it the referrals, and gave that commission its own inspector. A royal commission examined the programme with full inquiry powers. The public service commission ran a code of conduct investigation. The federal police opened a criminal investigation. Four separate mechanisms engaged with the same conduct and three of them produced adverse findings. The gap is at the end of each track rather than at the start of any of them.

The first reason it holds is that every consequence-bearing instrument requires either a higher standard of proof or a serving employee, and the record supplied neither at the point it mattered. A finding of serious corrupt conduct was made and then assessed as not provable beyond reasonable doubt, which leaves the finding standing and the prosecution unavailable. A finding of 97 code breaches was made and the two officials who accounted for a quarter of them had already left the service, which leaves the finding standing and the sanction unavailable. In both cases the investigation succeeded and the instrument it was supposed to feed had nothing to receive it.

The second reason is that the decision not to escalate is made inside the body that investigated, and no opened source identifies anyone who reviews it. The assessment that evidence would not sustain a prosecution was published by the same commission whose investigation produced the evidence. There is no visible mechanism by which that assessment can be tested, and no threshold that would make declining to refer a decision anybody has to defend. The one time in this whole record that an internal decision was successfully challenged, it took a statutory inspector, a misconduct finding, an external reconsideration by a former judge and eighteen months to reverse a refusal to investigate — and that reversal produced an investigation, not a consequence.

The third reason is that delay does the work quietly. Nine to eleven years separate the conduct from the findings, and thirty-two months separate the referrals from the outcome. Over an interval that long the people at the top of an administration retire, and retirement is the condition that removes the disciplinary instrument entirely. Nobody has to decide that a former secretary will not be sanctioned. The passage of time decides it, and the decision leaves no record because it was never made.

The fourth reason, which is what keeps the pattern stable rather than merely disappointing, is that the compensation track makes the whole thing look answered. More than 2.4 billion Australian dollars has moved, a court has approved a settlement for about 125,000 people, and the government has publicly accepted that the programme was unlawful. That is a great deal of visible response. None of it depends on anyone being held responsible, and none of it is reduced by nobody being held responsible, so the pressure that would otherwise build against the accountability gap is discharged somewhere else.

What observation would mean it is solved?

Candidates — (a) a prosecution is commenced, or a disciplinary sanction is imposed, against a person named in this record (b) a rule is put in place that attaches a consequence to conduct of this kind and can reach a person who has left the service, together with an external check on a decision not to refer a corruption finding to prosecutors (c) across corruption findings generally, the share that result in a prosecution or a sanction becomes measurable and rises.

(a) is the most direct test and it is now the least likely thing on the list. Every track that could produce it has formally closed. The criminal investigation ended in 2024, the code of conduct investigation ended in 2024, and the corruption investigation ended in March 2026 with an explicit assessment that the evidence would not carry a prosecution. Reopening any of them would require new evidence or a new decision maker, and no source opened here reports either being sought. Read as a live indicator, this one mostly measures whether something unexpected happens.

(b) is the observation that would show the pattern had actually changed, because it addresses the two conditions that produced this outcome rather than the outcome itself. Every response in this record has been an inquiry, and an inquiry ends when it reports. A standing rule about what follows a finding, and a standing check on the decision not to escalate one, would still exist after the attention has moved on. It is also the least likely to be adopted, because it would create a category of failure that the present arrangement does not have — a decision not to prosecute would become a decision someone could be found wrong about.

(c) alone counts ratios and can improve without anything improving. A rate rises when the denominator falls as easily as when the numerator climbs, so a commission that investigates fewer matters, or that makes findings only where the evidence is already strong enough to prosecute, would show a better ratio while doing less. The measure also has no baseline here: no source opened in this round gives a referral-to-prosecution rate for corruption findings in Australia, so there is nothing to compare a future figure against. The three have to be read together, and (b) is the one that would say whether the arrangement had changed rather than the weather.

What is it connected to?

Fills with researchaccountability for automated administrative decision systems in other jurisdictions, the operation of anti-corruption commissions in the Australian states as a comparison for the federal design, post-employment sanction regimes for former senior officials, the implementation record of the 57 Royal Commission recommendations, and the relationship between class action compensation and individual responsibility in public administration. Relation type and evidence grade were not confirmed in this round.

What these sources do not say

• How many people were issued a debt under the scheme. One opened source gives about 450,000 as an estimate. Research recorded other coverage using values between 433,000 and 526,000, and none of those was traced to a source that could be opened. No opened source states a counting basis, so whether the figures differ over withdrawn debts, over repeat debts against the same person, or over the boundaries of the scheme period cannot be determined.
• Whether any of the other ten officials found in breach was sanctioned. The opened reporting on the code of conduct investigation names the two former secretaries and states why neither could be sanctioned. It says nothing about the remaining ten of the 12, and no other opened source addresses them. Whether the finding of 97 breaches produced any disciplinary consequence at all is therefore unknown from this record, and the absence of that answer is the largest single hole in it.
• The reasoning behind the assessment that the evidence would not sustain a prosecution. Every account of that conclusion here is secondary. The investigation report itself did not open, so what evidentiary difficulty was identified, and whether it applies to both adverse findings or to one, cannot be read.
• What has become of the two people found to have engaged in serious corrupt conduct. No opened source published after March 2026 addresses their current position, employment, entitlements or standing. This appears to be an absence of follow-up reporting rather than a subject the sources declined to cover.
• Whether the honour awarded to one of the officials was revoked. Reporting from 2023 and 2024 records that revocation was called for. No opened source confirms whether it occurred, and the process sits with an office that was not examined in this round.
• Whether the assessment that no further recommendations were needed has been tested. The commission concluded that the Royal Commission recommendations already made were sufficient to prevent a repeat. No opened source records any independent body evaluating that conclusion, and whether such an evaluation exists or simply falls outside what was searched cannot be separated here.
• The text of the sealed chapter. The 56-page chapter reaches this document only through reporting of it. The tabled document itself was not opened, so the wording of the referrals and the reasons given for each are not available here.
• How the settlement figures fit together. The compensation of 475 million Australian dollars and the costs of up to 13.5 million are reported alongside a package figure of 548.5 million. The two named components sum to 488.5 million. No opened source names what accounts for the remainder.
• Almost the entire primary record, because it did not open. Repeated automated requests to the anti-corruption commission and Royal Commission pages timed out, and one legal centre summary returned a refusal. Four rows below are blank for that reason. Where a secondary account and an official document could differ in scope or wording, this document has no way to detect it, and the findings graded medium above are the ones that depend on that.

See the evidence

One official source opened directly and it is not the central one. The federal police media statement is the only government document here read at its own address, and it carries the closure of the perjury investigation and the reason given for it, which is why that row is the single high-grade item in the findings table apart from the absences. Every page of the anti-corruption commission and of the Royal Commission attempted in research timed out, so the two documents this problem is actually about — the sealed chapter of July 2023 and the investigation report of March 2026 — reach this document only through reporting of them. Four rows above are blank with the reason recorded rather than filled with a substitute link, and the load-bearing claim of the whole dossier sits on one of those blanks. Where sources overlap they agree: the split of two adverse findings against four clearances appears identically in two opened sources, as does the absence of any referral to prosecutors, and the code of conduct totals of 12 officials and 97 breaches appear in one opened source that also supplies the reason no sanction followed. Where the record does not add up the discrepancy is left visible rather than resolved — the settlement components of 475 million and up to 13.5 million sum to 488.5 million against a reported package of 548.5 million, and the population estimate of about 450,000 sits inside a spread of 433,000 to 526,000 that research observed and could not trace, so neither the sum nor the spread is silently normalised. The disciplinary outcome for ten of the 12 officials found in breach is absent from every source opened, and that absence is recorded as a finding rather than filled. This is a Path A output, so observation_refs is empty and provenance_mode: press-derived.

People affected

Estimated range 450,000–450,000 As of 2026 estimate of the 2015 to 2019 scheme period

Sensitivity The interval has zero width and is not a confidence interval. It is a single reported estimate, and the body that ran the scheme published no count that could be opened in this round. Research recorded other coverage using values from 433,000 to 526,000 for the same population, and no individual report carrying those values was opened, so the spread is stated here rather than used as bounds. A spread of that size, roughly plus or minus 10 percent around the figure used, is the honest uncertainty on this number even though the recorded interval does not express it. Two corrections would move the count in opposite directions and neither could be computed from what was opened: downward, because no source states whether debts later withdrawn are included or whether one person pursued more than once is counted more than once; upward, because the count covers people issued a debt and not the households behind them. The settlement registration count of about 125,000 is not used as a lower bound, because registering as a claimant is an act a person has to take and the population pursued is not limited to those who took it. The accountability quantities that this problem is actually about are several orders of magnitude smaller and answer a different question. The Royal Commission referred six individuals for corruption investigation on 2023-07-07. The Australian Public Service Commission reported on 2024-09-13 that 12 current and former officials had breached the public service code 97 times. The National Anti-Corruption Commission published on 2026-03-11 that two of the six had engaged in serious corrupt conduct and that the other four were cleared, and it referred neither of the two to prosecutors, on an assessment that the evidence would not prove the conduct beyond reasonable doubt. No criminal charge has been laid. The disciplinary side of that count is partly unavailable, because two of the 12 were beyond sanction after leaving the service and no opened source states the outcome for the other ten.

Regional breakdown The scheme was a Commonwealth programme administered nationally, and no source opened in research gives a state or territory count of people issued a debt under it. Splitting the national figure by resident population would be proportional allocation, and it would be wrong in both directions at once, because welfare recipiency is not distributed with population. The accountability process is federal throughout, so a geographic split would in any case not divide the failure this document measures.