From 10 December 2026 (OAIC, 2026), Privacy Act automated decision making rules require your privacy policy to say when software makes, or substantially helps make, decisions that significantly affect people. The new Australian Privacy Principle 1.7 asks for the kinds of personal information used and the kinds of decisions made. It applies to decisions after that date, including on data collected earlier.
Drafting the paragraph is quick. Finding every automated decision across your systems and vendor tools is what takes weeks.
Key takeaways
- APP 1.7 commences on 10 December 2026 and applies to any APP entity that uses a computer program, with personal information, to make or substantially support decisions that could significantly affect a person.
- Decisions count whether the effect is adverse or beneficial: credit, insurance, employment, contracts and access to significant services are all in scope.
- A person who only confirms what the system already decided does not take the decision out of scope.
- The organisation that uses a vendor's tool carries the disclosure obligation, so supplier information requests need to start now.
- Sequence the work: inventory, assess, fix weak controls, then disclose.
Is AI regulated in Australia if there is no AI Act?
Most attention on AI regulation in Australia has focused on what did not happen. The mandatory guardrails floated in 2024 were set aside, and the December 2025 National AI Plan relied on voluntary guidance instead. That left a common impression that AI in business is largely unregulated.
It is regulated, through existing law. The Privacy and Other Legislation Amendment Act 2024 (Cth) added new transparency obligations to the Privacy Act 1988 (Cth), commencing on 10 December 2026. From that date, any organisation covered by the Privacy Act that uses a computer program to make or substantially support decisions affecting individuals must say so, in plain terms, in its privacy policy.
The rule captures lenders assessing applications, insurers pricing premiums, employers screening applicants, and any organisation that has wired an AI model or automated rules engine into a customer-facing process. If you have not yet mapped where automated decisions happen, the window to prepare is closing.
The regulation you were waiting for is already here The absence of a standalone AI Act does not mean AI is unregulated. The Privacy Act, the Corporations Act, anti-discrimination law and consumer law all already apply to automated systems. The 2024 amendment makes one of those obligations, transparency, explicit and enforceable.
What does APP 1.7 require?
The amendment expands Australian Privacy Principle 1, which governs what an organisation must put in its privacy policy. The new APP 1.7 requires an APP entity to include information about automated decision making where it uses personal information in a relevant automated process.
The trigger is specific. The obligation applies where an APP entity arranges for a computer program to use personal information about an individual to make a decision, or to do a thing substantially and directly related to making a decision, that could reasonably be expected to significantly affect the individual's rights or interests. Where that test is met, the privacy policy must set out:
- the kinds of personal information used in the operation of the automated decision-making technology, and
- the kinds of decisions made solely or substantially by that technology.
A companion provision, APP 1.9, explains "significantly affect". A decision counts whether it affects the individual adversely or beneficially. Examples include a decision under an Act to grant or refuse a benefit, a decision that affects an individual's rights under a contract or arrangement, and a decision that affects access to a significant service or support. That breadth means many routine commercial decisions are covered.
| Likely in scope | Likely out of scope | Impact | |
|---|---|---|---|
| Lending | Automated credit approval or decline | Internal fraud-flag review by a human | In scope |
| Insurance | Algorithmic premium or eligibility decision | Aggregate actuarial modelling | In scope |
| Employment | AI shortlisting or ranking of applicants | Roster optimisation with no personal impact | In scope |
| Services | Automated eligibility for a benefit or support | Product recommendation with no rights impact | Assess |
Does a human in the loop take a decision out of scope?
The distinction that matters most is between a decision a program makes or substantially supports, and one where a person exercises real judgement using a tool. A model that scores an applicant, followed by a person who rubber-stamps the score, sits much closer to automated decision making than the presence of a person suggests. The OAIC has signalled that meaningful human involvement, as opposed to token sign-off, is what separates the two.
Why is the disclosure only the visible part of the work?
The privacy policy wording is the smallest part of the job. Writing two paragraphs is easy once you know what they should say. The hard part is that most organisations do not have a current, accurate inventory of where automated decisions are made.
Decision logic builds up quietly: a rules engine added to a loan origination system years ago, a third-party scoring API on an onboarding flow, a recruitment platform whose ranking feature was switched on by default, a pricing tool a vendor upgraded to include a model. Each was a sensible choice. Together they mean the honest answer to "where do we make automated decisions about people?" is often "we are not sure". You cannot disclose what you have not mapped.
A defensible disclosure needs an AI and automation inventory, an assessment of which entries meet the significant-effect test, and a view on whether human involvement is meaningful in each. That work also surfaces discrimination exposure in automated screening, consumer law exposure where automated decisions mislead, and governance gaps the board will want to know about.
Automated decisions tend to sit in three layers: systems the organisation commissioned, AI features switched on inside existing software, and informal automations built by teams in low-code tools or spreadsheets. Each layer is harder to see than the last.
| Decision source | Typical visibility to a privacy officer | Impact | |
|---|---|---|---|
| Commissioned decision engine (lending, pricing) | Owned by one team | Documented but rarely mapped to privacy policy | Medium |
| AI feature inside a SaaS product | Vendor-controlled | Often undisclosed in contracts and unmonitored | Low |
| Low-code or spreadsheet automation | Built by a team locally | Usually invisible to central governance | Very low |
| Third-party scoring or enrichment service | Embedded via API | Data flows poorly understood | Low |
Organisations that already keep an AI register, as described in our AI governance framework guide, are well ahead. A useful register records decisions instead of systems: approve a loan, shortlist a candidate, set a premium. For each one, capture the owning team, the software involved, the kinds of personal information used, and whether a person meaningfully intervenes.
- InventoryMap every automated or AI-assisted decision
- AssessWhich meet the significant-effect test
- Review controlsHuman involvement, fairness, records
- DiscloseUpdate the privacy policy in plain English
How does the APP 1.7 test work in practice?
For example, consider a hypothetical Australian non-bank lender. Applicants enter income, expenses and identity details online. A scoring model returns a risk grade. The origination system automatically declines applications below one threshold, auto-approves those above a higher one, and refers the middle band to a credit officer.
Two of those three paths are automated decisions in the APP 1.7 sense. The auto-decline and auto-approve are made solely by the program, use personal information, and plainly affect the applicant's rights under a prospective contract. The referred band is closer to human decision making, provided the credit officer exercises real judgement instead of deferring to the score.
From 10 December 2026 that lender's privacy policy must describe the kinds of personal information the model uses and state that some lending decisions are made solely by automated technology. Getting there means documenting the model's inputs, confirming which paths are fully automated, and deciding whether the referred path counts. That is systems mapping, and the legal drafting depends on it. The same discipline sits behind good AI advisory work for boards.
Swap the lender for an insurer setting premiums, a labour-hire firm ranking candidates or an aged care provider running eligibility checks, and the analysis is the same. Aged care providers will recognise the overlap with AI automation under the new Aged Care Act, where automated eligibility for support is exactly the kind of "significant service" decision APP 1.9 describes.
What are the penalties for getting it wrong?
The transparency requirement arrives with stronger enforcement powers. The 2024 reforms gave the OAIC new tiers of civil penalty and the ability to issue compliance notices and infringement notices for specified contraventions, including privacy policy shortfalls. A privacy policy that does not meet the automated decision-making requirements is the kind of gap those notice powers address without full litigation.
For a body corporate, the maximum civil penalty for a serious interference with privacy is the greatest of $50 million, three times the benefit obtained, or 30% of adjusted turnover for the relevant period (Privacy Act 1988, as amended in 2022). Infringement notice amounts are fixed in the legislation and are much lower, but they can be issued quickly.
- A complete inventory of automated decisionsOne list, owned
- Early sight of discrimination and fairness riskBefore a complaint
- A privacy policy that survives regulator scrutinyWritten to the APP 1.7 test
- Board-level assurance on AI decisioningGovernance ready
Reputational exposure is often larger than the direct penalty. Disclosing accurately that you make automated lending or eligibility decisions invites questions about fairness. That is far safer than being found to have run undisclosed automated decisions that affected people, then having it surface through a complaint, a journalist or the regulator. The more immediate risk for most organisations is an OAIC inquiry that exposes undocumented automated decisions, followed by remediation under scrutiny.
Does APP 1.7 apply to your organisation?
Two thresholds decide it. First, is the organisation an APP entity? That covers businesses with annual turnover above $3 million, plus organisations under that threshold in health, credit reporting and some related fields. Second, does it make automated decisions that meet the significant-effect test? The guide below is a starting filter, not legal advice.
Do you use a computer program, with people's personal information, to make or substantially drive decisions about them?
- Yes, and those decisions affect rights, contracts or access to servicesAlmost certainly in scope: map and disclose
- Yes, but a person exercises genuine judgement each timeAssess whether human involvement is meaningful
- Automated, but no material effect on individualsLikely out of scope: document why
- Not sure whether any decisions are automatedStart with an inventory: you cannot rule it out
The most dangerous answer is the last one. An organisation that cannot say whether it makes automated decisions is still covered, and exposed without knowing it. Financial services firms have the clearest exposure, but retailers, labour-hire firms, health providers and membership organisations should all check.
How do you get ready by 10 December 2026?
The work divides into discovery, assessment, remediation and disclosure. Discovery is the phase organisations underestimate. Mapping automated decisions across several systems and third-party tools takes weeks, because the knowledge is spread across teams and some of it lives only in vendor configurations.
- Phase 1DiscoverInventory every system and tool that makes or supports decisions about people
- Phase 2AssessApply the significant-effect test and check human involvement
- Phase 3RemediateFix fairness, records and human-review gaps before disclosing
- Phase 4DiscloseUpdate the privacy policy and brief the board before 10 December
Sequencing matters. Disclosing before remediating means publishing an accurate description of practices you have not yet made defensible. Tightening human review, documenting model inputs and adding fairness checks should happen before the practice becomes public record.
If you are weighing outside help, a good partner speeds up discovery, brings a tested assessment framework and leaves you understanding your own systems. A poor one produces a document and leaves the decisions as opaque as before. Our guides on how to choose an AI consultancy in Australia and the AI consulting buyer's guide apply directly.
What mistakes turn APP 1.7 into a scramble?
- Assuming a person in the process takes the decision out of scope. If their role is to confirm what the system decided, the decision remains substantially automated. Token review is exactly what regulators have signalled they will look behind.
- Treating vendor tools as the vendor's problem. The organisation that deploys a ranking platform, credit-scoring API or pricing engine to decide things about people carries the obligation. Older contracts rarely give you the information you now need, and getting it from suppliers takes time.
- Disclosing everything out of caution. Over-disclosure invites scrutiny of processes that are not automated decisions and dilutes the ones that matter. A precise disclosure is more compliant and more defensible.
- Leaving the board out until the end. Decisions that significantly affect people are a governance matter. The mapping should reach the board well before the deadline, as assurance instead of surprise.
How do you write an APP 1.7 disclosure?
APP 1.7 asks for the kinds of personal information used and the kinds of decisions made. "Kinds" matters: you describe categories, and you do not publish the model or its logic. A lender does not disclose its scoring algorithm. It discloses that it uses information such as income, expenses and credit history to make some lending decisions automatically.
Plain English matters more than legal precision. The audience is the affected individual. The test is whether a customer reading the policy would understand that automated decisions are made about them, what information feeds them and what kinds of decisions they are.
| Weak disclosure | Strong disclosure |
|---|---|
| "We may use automated systems to process your information." | "We use an automated system to assess applications for credit. It uses your identity details, income and expense information, credit history and transaction data to produce a decision or recommendation. You can ask us to explain a decision and request human review." |
Which sectors are most exposed to APP 1.7?
- Financial services and lending: credit decisioning, affordability assessment and fraud scoring are heavily automated. These firms also answer to APRA and ASIC, so disclosures must be consistent across regulators, covered in financial services AI compliance under APRA and ASIC.
- Insurance: underwriting, pricing and claims triage increasingly use automated models. A declined claim or sharply higher premium is exactly what the reform targets.
- Property and real estate: tenancy screening platforms rank and filter applicants, often with little human reconsideration, affecting access to housing.
- Recruitment and large employers: applicant tracking systems that score and shortlist candidates affect employment and use personal, sometimes sensitive, data.
- Health and allied health groups: any automation affecting access to care, eligibility or prioritisation clears the significance threshold.
What should the board ask before December?
- Do we have a complete register of the automated decisions we make that significantly affect individuals?
- For each one, do we know what personal information the system uses?
- Can a person ask for, and receive, a human review of an automated decision that affects them?
- Are we confident the decisions are accurate and free of unlawful discrimination?
- Is our privacy policy specific enough to meet the new obligation, or is it generic boilerplate?
If the honest answer to any of these is "not yet", that gap is the work between now and the deadline.
The automated decision-making rules sit alongside the National AI Plan, the National AI Centre's Guidance for AI Adoption and directors' existing duties under the Corporations Act. Treat 10 December as the first concrete instance of a broader expectation: being able to explain and defend how software makes decisions about people. Our Australian AI rules and deadlines tracker keeps the related dates in one place.
How Solve8 can help
Our team maps every automated or AI-assisted decision across your systems and supplier contracts, applies the significant-effect test, flags the controls to fix first, and drafts plain-English disclosure wording for your lawyer or privacy officer to approve. The work runs inside your own systems as a fixed fee per phase, agreed before we start, and you can stop after the first phase with the register in hand. See our AI strategy services or talk to us.
