Summary
- The Notifiable Data Breaches scheme puts two decisions in the hands of the organisation that holds the data: whether it has grounds to suspect an eligible data breach, and whether a reasonable person would conclude the incident is likely to cause serious harm.
- The regulator publishes statistics on notifications received. In the retained reading of its statistics page on 17 November 2026 there is no count of assessments that ended in no notification and no denominator of covered entities, and the reading itself warns that the live dashboard's table structure can change.
The part, the scheme, and the guide
The retained summary of the Privacy Act 1988 describes Part IIIC as the part that sets out requirements for entities to notify eligible data breaches, and names the Notifiable Data Breaches scheme as the scheme that part establishes; section 26WA is described as the guide to the part. That much is orientation. What follows is machinery.
Across both retained snapshots of 17 November 2026 — a provider summary of the Privacy Act 1988 held by the Federal Register of Legislation, and a provider summary of the regulator's statistics page — each carried a null publisher excerpt. No publisher-verbatim sentence was held, so nothing below is quoted and no statutory wording appears in quotation marks.
The clock starts with the entity's own awareness
section 26WH(1) is the provision that makes the scheme run, and where it starts is the point. The retained summary describes it as obliging the entity to carry out a reasonable and expeditious assessment of the suspected eligible data breach within 30 days after the entity becomes aware that there are reasonable grounds to believe there may have been an eligible data breach.
Not 30 days from the intrusion. Not 30 days from the day the logs were read. The period attaches to the entity's own state of awareness, and the threshold for that awareness is a belief about grounds — grounds to believe a breach may have occurred. A belief that grounds may exist is a long way from proof, which is why the assessment exists at all. It is also a state that no outsider observes, records or certifies. That is the first selection point, and the selecting party is the one being counted.
The text is held as a reported reading, not as bytes: see the Privacy Act 1988 on the Federal Register of Legislation.
Serious harm is a forecast, not a finding
section 26WE carries the second selection point. The retained summary describes a two-limbed test: unauthorised access to, unauthorised disclosure of, or loss of, information held by an entity; and a reasonable person would conclude that the access or disclosure would be likely to result in serious harm to any individual to whom the information relates. Where the event is a loss, the summary describes a separate route — whether access or disclosure is likely to occur as a result of the loss, and whether a reasonable person would then conclude that such access or disclosure would be likely to result in serious harm.
Every operative verb there points forward. Likely to result. Would conclude. The scheme is not asking whether harm occurred; it is asking what a reasonable person would predict about harm that has not yet happened. Predictions made in the middle of an incident, by the party whose conduct and reputation are on the line, on facts it collected and framed, are the raw material of the count. A forecast can be wrong in either direction. Only the entity holds the file from which its forecast could be examined.
What has to be handed over, and to whom
section 26WK is described as requiring a statement containing specified information — the entity's identity, a description of the breach, the kinds of information concerned, and recommendations about the steps individuals should take — and requiring that statement to be provided to the Commissioner. section 26WL is described as setting out the notification requirements, reaching both the Commissioner and affected individuals.
Two addressees, one document set, one decision generating both. A reader who wants to test whether a particular incident was classified correctly is not short of a recipient to ask. What the reader cannot obtain is the reasoning that produced the classification, because the reasoning is not the product being filed.
The rest of Part IIIC was not read: no penalty provision, no definition of the entity or of serious harm, no exemption route, no provision permitting or requiring an entity to record a decision that an incident is not eligible, and no duty on the Commissioner to publish anything under that part. Nothing is asserted here in either direction about any of them.
What the published table lets an outsider check
On 17 November 2026 the retained reading of the regulator's notifiable data breach statistics page described it as publishing statistics through a dashboard and periodic reports, with the reports disclosing the number of notifications received. The statistics table is described as typically covering fields such as the source of the breach — malicious or criminal action, human error and system fault given as examples — and the sectors involved.
The same reading describes the page as not publishing a comprehensive count of non-eligible incidents, and as not publishing a standardised denominator or rate for the regulated population. Its own uncertainty field notes that the specific current table structure is subject to change in the live dashboard. The reading names periodic reports and gives no interval, so no cadence — monthly, quarterly, annual or otherwise — is asserted here.
No report, dashboard, table, period total, sector share or per-notification figure was inspected for this article. It therefore quotes no number of notifications at all, and its silence should not be read as a figure of zero.
What the described fields do carry is the entity's own account of the incident's cause, sorted into buckets that the entity filled. A rise in the human error bucket is a statement about how organisations labelled their own failures as much as about how they failed. A table with no set-aside column cannot distinguish a quiet quarter from a quiet filing habit, and a table with no covered-population figure cannot turn any total into a rate.
The delta, measured rather than claimed
Two archive searches run on 17 November 2026 at a limit of 25 — one for the scheme's own vocabulary, one for the regulator's acronym — returned no rows, while the directory entry for the Office of the Australian Information Commissioner in our directory records one earlier article attached to it. That is the measurement this piece tests, and it is not a claim of firstness.
What the count is good for
The published number is a real record of something: how many organisations, in a period, decided to hand an incident over. That is worth counting. It is not a census of breaches, and no arithmetic applied to it can make it one, because the population it is drawn from is selected before the counting begins by the parties being counted, at two separate moments, using a predictive standard they apply to their own facts. A board that treats the figure as its exposure has mistaken a record of disclosure decisions for an inventory of incidents. The control that closes the gap is not analytical.
It is documentary: a dated record of when awareness arose, who made the harm call, and what the entity examined and set aside.
Member Briefing
Read the full trend briefing
Members read the full briefing with the source notes behind every finding.
Strategic Circle
A daily briefing in your language, built from changes we have verified.
Join for USD 20/monthCancel anytime.
Members also get the full research library, a daily briefing on the networks they follow and the private forum.






