Compliance reference · OAIC Privacy APPs
The Privacy Act 1988 and the thirteen Australian Privacy Principles set the rules for handling personal information in Australia. AI deployments expand the collection footprint in ways that are easy to miss. This page is the plain-English version, and where AI changes the calculus.
What it actually requires
The thirteen Australian Privacy Principles do not all feel the same in practice. The four groupings below are how most operations teams actually think about the obligations day-to-day.
APPs 1–5
Have a clear privacy policy. Allow individuals to interact anonymously where reasonable. Collect personal information only when reasonably necessary. Notify individuals about collection. Apply solicited and unsolicited information rules.
APPs 6–8
Use and disclose personal information only for the purpose it was collected for, or for related purposes the individual would reasonably expect. Direct marketing has its own rules. Cross-border disclosure carries explicit accountability for the receiving party.
APPs 10–13
Keep personal information accurate, current, and complete. Take reasonable steps to secure it from misuse, interference, loss, and unauthorised access. Allow individuals to access and correct their information. Government-related identifiers have their own constraints.
NDB scheme
Eligible data breaches likely to result in serious harm must be assessed promptly, contained, and notified to the OAIC and affected individuals within the prescribed window. The notification is structured — what happened, what information was involved, what to do about it.
Four problems that show up when AI lands inside an existing privacy program.
Customer files in CRM, transcripts in a recordings platform, copies in shared drives, exports in someone's downloads folder. When an access request lands the team scrambles to assemble what the individual is entitled to see.
You cannot prove what you hold or what you have deleted.
Cloud AI tools cache transcripts, train on prompts, and retain logs. Collection happened the moment the prompt was sent. Whether the individual would reasonably expect that use is a question the audit will ask.
Your AI deployment expanded your collection footprint.
Something happened. Someone asks "is this notifiable?" and the team relies on shared memory of past incidents. The assessment record, the harm test, the containment timeline — none of it is structured.
Your assessment cannot be reproduced for the OAIC.
Each access request requires gathering everything held about the individual across systems, redacting third-party information, and responding inside the timeframe. Manual collation is slow, inconsistent, and prone to omission.
Each request is a fresh fire drill.
How BackPro maps to the APPs
The aim is for AI use to fit inside the privacy boundary the firm already operates — not to invent a parallel one. Each row maps a privacy obligation to the part of BackPro that does the work.
Privacy obligation
Privacy policy and collection notice text live in the platform with version history and approval signatures. Updates trigger workflow on dependent surfaces — websites, application forms, contact-record templates — so the public-facing text and the internal record stay in step.
Privacy obligation
AI prompts and outputs run through a two-tier semantic gate before reaching the user. The gating layer can refuse outputs that exceed the documented purpose, downgrade them, or flag them for human review. Use-and-disclosure decisions become enforceable, not aspirational.
Privacy obligation
BackPro deploys inside your tenant with encryption at rest and in transit, role-based access control, and tenant isolation. The audit log signs every access with HMAC-SHA256 and chains it to the previous entry. The Control Monitor agent watches for control drift and stale evidence.
Privacy obligation
Subject access requests run as a workflow with full chain of custody — what was searched, what was returned, what was redacted, who approved the response. The audit log produces evidence of completeness without retrospective reconstruction.
Privacy obligation
Breach assessment runs as a structured workflow — incident facts, information types involved, harm assessment, containment actions, decision rationale. Each step is signed and chained. The OAIC notification is generated from the assessment record, not assembled from email threads.
Privacy obligation
Models run inside your tenant. Prompts and outputs do not leave the perimeter. Source attribution and the gating layer keep AI use within the boundary the privacy policy describes — not whatever the upstream LLM happens to support.
Privacy accountability is broader than CPS 234 — three desks share the load.
What APP entities ask before bringing AI inside the privacy boundary.
One walkthrough covers architecture, audit chain, deployment model, and how the platform maps to the privacy obligations that apply to your firm.