Privacy Act automated decision-making transparency (APP 1.7 to 1.9)

What it is

The Privacy and Other Legislation Amendment Act 2024 received Royal Assent on 10 December 2024. Among other changes, it inserted three new subclauses into Australian Privacy Principle 1 of the Privacy Act 1988. APP 1.7, 1.8 and 1.9 create a transparency duty for automated decision-making (ADM), and they commence on 10 December 2026.

On 30 September 2026 the OAIC released its final guidance. It updated chapter 1 of the APP Guidelines (version 2.0) and published an APP 1.7 to 1.9 fact sheet, a flowchart and a supplementary fact sheet for government agencies. The OAIC says the resources reflect 90 written submissions to its consultation.

Who it applies to

Every APP entity. That covers Australian Government agencies and private sector organisations with annual turnover above A$3 million, plus the smaller businesses the Act already brings in, such as health service providers. The small business exemption still applies. The government has said its removal belongs to a later tranche of reform, and nothing to remove it has been legislated.

The obligation falls on the entity that has “arranged for” the computer program to be used. The OAIC expects it to stay with the organisation using the personal information to make the decision, even when a vendor operates the system.

What it requires

APP 1.7 is triggered when three conditions are met.

• The entity has arranged for a computer program to make a decision, or to do a thing substantially and directly related to making one.
• The decision could reasonably be expected to significantly affect the rights or interests of an individual.
• Personal information about the individual is used in the program’s operation.

When it is triggered, APP 1.8 says the privacy policy must contain three things.

• The kinds of personal information used in the operation of those programs.
• The kinds of decisions made solely by the operation of those programs.
• The kinds of decisions for which a thing substantially and directly related to making the decision is done by those programs.

A decision includes refusing or failing to decide. It can affect someone beneficially or adversely. A human decision that is merely recorded in software, such as in a word processor, is not caught.

The OAIC allows entities to group decisions and personal information into categories. Grouping is not a licence to be vague. The description must be meaningful to a reasonable person, and the policy should make clear where more sensitive information is used. There is a narrow exception for commercially sensitive information. It does not extend to information that merely has commercial value.

The scope test, made concrete

Most organisations get the scope question wrong because they ask “where do we use AI”. The law asks where a computer program shapes a decision about a person. Work through two tests for every system that touches such a decision.

Does a computer program make or substantially shape the decision?

The OAIC gives “computer program” its ordinary meaning and reads it broadly. It names rule-based processes, AI, machine learning and generative AI tools such as chatbots. In practice that catches credit and risk scores, rules engines that approve or decline applications, eligibility checks, automated refunds and third-party tools that shortlist, price or flag people.

“Substantially” means real and of substance, not nominal. “Directly” means a direct connection to the decision. An advisory output can qualify. The OAIC’s examples include a loan risk score that officers almost always follow and a suitability score built into the screen an assessor works from. The more a person relies on the output, the more likely it is in scope.

Could the decision significantly affect the person?

“Significantly” means more than trivial, with the potential to considerably influence a person’s circumstances or outcomes. Rights include legal, social and economic rights. Interests include liberty, livelihood, employment, health and finances. The OAIC’s examples include granting or refusing a benefit, decisions under an insurance contract, access to healthcare, banking, utilities, housing or employment, and targeted pricing for significant goods or services. Context matters. A decision can be significant for a child or a person experiencing vulnerability when it would not be for others.

Routing an enquiry to the right queue, de-duplicating records or a dashboard that informs staff without deciding anything about a person will usually fall outside.

Sorting the results

Put each system into one of three piles. Flag systems that clear both tests. Set aside systems that clearly fail one, with a one-line reason. Mark close calls grey, record your reasoning and treat them as in scope until you have a basis to move them out. The OAIC’s own message is not to take too narrow a view.

The common grey cases are a human who signs off the output (ask whether they genuinely weigh it or mostly endorse it), a fraud flag that sends a case to manual review (closer to set aside, unless the flag restricts the person while they wait), a ranking that a person then acts on (judge it by influence on the result) and a low-stakes decision applied to a large group.

What to have in place by 10 December 2026

• A list of the systems in scope. You cannot describe the kinds of decisions a program makes if you do not know which programs make them. Walk each team, include third-party tools and spreadsheets with decision logic, and run the two tests.
• The facts the disclosure rests on. For each flagged system, the decision it makes or shapes, whether it decides alone or supports a person, the personal information it uses and whether any of it is sensitive.
• An updated privacy policy. Grouped, plain descriptions of the kinds of decisions and the kinds of personal information, for both solely automated and substantially assisted decisions.
• Sign-off. Your privacy officer, legal counsel and in many cases the board need to agree what is disclosed. This is often the slowest step.
• A way to keep it current. The obligation applies to decisions made after commencement, including by systems set up years earlier. New systems arrive and existing ones change, so the triage has to be rerun whenever a system goes live or changes materially.

Penalties and enforcement

The OAIC can issue infringement notices where a privacy policy is not clear and up to date (APP 1.3) or lacks required content (APP 1.4). Each contravention carries up to 60 penalty units for a body corporate, or 200 for a listed corporation. It can also issue compliance notices. For the most serious or repeated interferences with privacy, a body corporate faces civil penalties of up to the greater of A$50 million, three times the benefit obtained or 30% of adjusted turnover for the relevant period.

How it relates to other frameworks

APP 1.7 makes part of the transparency principle in Australia’s AI Ethics Principles a legal duty. It is a disclosure rule, not a right to object, which makes it narrower in effect than the GDPR’s Article 22 but wider in scope, because it is not limited to solely automated decisions. Commonwealth agencies apply it alongside the DTA policy and may cross-link their ADM disclosures with the Information Publication Scheme, as the OAIC suggests. The OAIC’s earlier guidance on AI products covers the rest of the APPs as they apply to AI.

Working with it in Aicura

The ADM (Privacy Act) framework is available from the Privacy tier. You upload your privacy policy and Aicura reviews it against the APP 1.7 to 1.9 requirements, using your AI Register as context, and returns recommendations tied to specific findings. Each new version of the policy is reviewed again, and the review history stays with the policy.

The AI Register holds the facts the disclosure rests on. Each entry records what the system produces (such as a recommendation or score, or an autonomous decision), whether it affects individuals, the level of human oversight, whether it processes personal information and the data categories involved. Discovery surfaces AI tools already in use that are not yet on the register, so the list you triage is closer to complete. From Essentials, a Decision Disclosure document can be generated and accepted for each entry.

Aicura provides guidance. It does not decide what your privacy policy must say or give legal advice.