The register is the public face of NDIS enforcement. It lists compliance notices, banning
orders, revocations of registration, refusals to re-register, suspensions and enforceable
undertakings, and anyone can download the whole thing as a spreadsheet. What it does not do
is show you what it looked like last month. Entries are not removed when they stop applying;
they gain a date in the Date no longer in force
column and stay on the list.
That makes the register a poor guide to how much enforcement is currently in force, and a very good one if you keep your own copies. We keep ours.
The two numbers move in opposite directions
| Counted as at its own download date | 19 Jul 2026 | 28 Aug 2026 | Change |
|---|---|---|---|
| Entries on the register | 3,671 | 3,734 | +63 |
| Actions in force | 2,468 | 1,885 | -583 |
| Compliance notices in force | 1,338 | 703 | -635 |
| Banning orders in force | 656 | 685 | +29 |
| Revocations of registration in force | 333 | 356 | +23 |
| Refusals to re-register in force | 139 | 139 | 0 |
Actions in force fell by 24 per cent in six weeks. Compliance notices in force fell by 47 per cent, close to half. And the whole of the decline is that one instrument: every other category held steady or rose.
Where the 639 went
Comparing the two files row by row, 639 entries that carried no end date on 19 July had one by 28 August. None went the other way. They closed out in batches:
| End date recorded | Notices closed |
|---|---|
| 4 August 2026 | 241 |
| 6 August 2026 | 237 |
| 28 July 2026 | 115 |
| 24 July 2026 | 21 |
| 15 July 2026 | 16 |
| Other dates | 9 |
They came from two cohorts: 375 had taken effect on 30 January 2026 and 240 on 30 March 2026. Those are the audit sweeps this masthead has written about before, seen from the other end. A sweep issues several hundred compliance notices on one day; some months later they come off in a similar batch. 638 of the 639 were compliance notices; one was a banning order.
This is a lifecycle, not a scandal, and we want to be plain about that. A compliance notice is a direction to fix something; it ending is consistent with the thing being fixed. Nothing in the register says why any individual notice ceased, and we are not going to guess.
Why it matters that the list still grows
What the arithmetic does is break a natural reading. If you look at this register in July and again in August, it is bigger, and bigger reads as more enforcement. In this six weeks the opposite happened to the amount of enforcement actually operating, by a quarter.
We made a related point on 10 August about the 2025 figures, that a fourfold rise in published actions was ninety per cent one instrument. This is the same distortion on a different axis. The count of entries is a record of everything the Commission has ever published; the count in force describes only the day the file was taken. They are different questions and only one of them is easy to get.
Our view, labelled as such. The serious end of the register is the part that rose. Banning orders in force are up 29 and revocations up 23 in six weeks, while the administrative instrument halved. If you want a single sentence about the direction of NDIS enforcement in this period, it is that there is less of it in force and more of it at the serious end, and neither half of that is visible from the register's own headline count.
A reading rule we have not published before
One more thing worth knowing before you cite a row. An entry can be on the register
before it takes effect. On 28 August five rows carried a Date effective from
later than that day: two revocations of registration and two banning orders dated 31 August,
and one revocation dated 4 September. On 19 July there were four such rows.
So a row appearing on the register is not the same as a sanction currently applying, in either direction: some have not started, and 1,844 of the 3,734 entries have already ended. If you are checking whether a provider is under a live sanction, both date columns matter.
How we did this
We downloaded the register's full CSV export on 19 July 2026 and again on 28 August 2026 and compared them in Python. Both files are kept. Counts of what is in force are taken as at each file's own download date: a row counts as in force if its effective date has arrived and its end date either is blank or is later than that date. Counting both files as at today would have flattered the July figure.
The comparison key decides the answer, and our first two attempts got it wrong in opposite directions. Matching rows on type, name, effective date and end date said 648 rows had vanished, which was false: rows that merely gained an end date looked like departures. Matching on type and name alone said 8. The reliable method is to match on the fields that identify the action, type, name and effective date, and then compare the end date within those matches. That is how the 639 figure is derived, and it is the number we would want checked first.
Genuinely departed rows, meaning entries present in July and absent in August, number 8 on the identifying key. We are not drawing a conclusion from them here: a register can drop a row for correction, duplication or review, and eight out of 3,671 over six weeks is not enough to characterise. We note it so the next comparison starts from a number.
We have not contacted the NDIS Commission. This is arithmetic on its own published export. Where we have drawn an inference, it is labelled as our view. No provider or person is named in this piece; the analysis is aggregate.
Sources
- NDIS Quality and Safeguards Commission, Compliance and enforcement actions register, full CSV export (downloaded 28 August 2026): 3,734 entries; the type, effective-date and end-date columns for every row; the five future-dated entries; and the 1,885 actions in force on that date.
- NDIS Quality and Safeguards Commission, the same CSV export downloaded 19 July 2026 and retained by this masthead: 3,671 entries and 2,468 actions in force on that date. The register publishes only its current state, so the July figures in this story cannot be reproduced from the live download and rest on our saved copy.