Your systems already make decisions about people. From 10 December you have to say so.

What actually commences on 10 December 2026

The Privacy and Other Legislation Amendment Act 2024 added three new paragraphs to Australian Privacy Principle 1. They commence on 10 December 2026, and they change what an organisation’s privacy policy has to say.

The trigger in APP 1.7 has three parts. An organisation has arranged for a computer program to make a decision, or to do something substantially and directly related to making it. The decision could reasonably be expected to significantly affect the rights or interests of an individual. And personal information about that individual is used in the operation of the program.

Where all three are true, APP 1.8 requires the privacy policy to contain three things. The kinds of personal information used in the operation of those computer programs. The kinds of decisions made solely by the operation of a computer program. And the kinds of decisions where a computer program does something substantially and directly related to making the decision, which is the more common case of a person deciding with a system shaping the outcome.

APP 1.9 gives examples of decisions that may significantly affect rights or interests. A decision under legislation to grant or refuse a benefit. A decision that affects someone’s rights under a contract, agreement or arrangement. A decision that affects their access to a significant service or support.

The obligation applies to APP entities. That generally means Commonwealth agencies and organisations with annual turnover above $3 million, plus some smaller businesses the Act brings in regardless. State and territory agencies sit under their own privacy laws, so a state department’s obligations need checking against its own jurisdiction.

None of this requires an organisation to stop automating. It requires the organisation to know what it has automated, and to say so.

Why this is not an AI problem

The instinct in most organisations will be to hand this to whoever owns the AI register. That is the wrong list.

The explanatory memorandum says “computer program” is intended to take its ordinary meaning and to cover pre-programmed rule-based processes as well as artificial intelligence and machine learning. The Office of the Australian Information Commissioner’s Issues Paper, published in May 2026, repeats that breadth. A rules engine written a decade ago is as much in scope as a model trained last month.

Consider what sits inside an ordinary enterprise estate. A recruitment module that scores and ranks applications before a person reads any of them. A labour-hire or contractor roster that ranks who is offered a shift. A customer platform that flags accounts for review, suspension or a change of terms. A credit or payment system that decides who is offered what. Very little of that would appear on an AI register. Much of it can meet the APP 1.7 test.

There is one significant qualification for workforce systems, and it deserves to be stated plainly rather than buried. The Privacy Act’s employee records exemption means that a private-sector employer’s handling of an employee record, where it relates directly to a current or former employment relationship, generally sits outside the APPs. How that exemption interacts with the new transparency obligation has not been settled by the regulator. What is settled is that the exemption does not reach job applicants who are not hired, volunteers or independent contractors, and it does not apply to Commonwealth agencies at all. Recruitment, contractor rostering and customer decisioning are therefore where most organisations should start.

The practical consequence is the same either way. An organisation cannot decide which of its automated decisions are exempt until it knows which automated decisions it makes.

The enforcement half nobody is writing about

Most commentary on APP 1.7 treats it as a disclosure nicety with a long lead time. It is neither.

The same 2024 Act gave the OAIC infringement notice and compliance notice powers, and those have been in force since 11 December 2024. Failures in a privacy policy are within their reach. The OAIC has said publicly that entities with non-compliant privacy policies may face compliance and infringement notices and penalties of up to $66,000.

It is also already looking. In the first week of January 2026 the OAIC began a compliance sweep of privacy policies, reviewing around sixty entities across six sectors against the existing APP 1 requirements. The sweep did not target automated decisions, because the new paragraphs had not commenced. It did establish that the regulator reads privacy policies, systematically and in public, and is prepared to act on what it finds.

From 10 December 2026, the automated decision-making disclosures become part of what a privacy policy is read against.

Why your programme will find this late

Here is the structural problem, and it is the reason this is a governance issue before it is a legal one.

Most automated decision capability in an enterprise did not arrive through a business case. It arrived through a vendor release. A new ranking feature in the recruitment module. A default rule in the rostering engine. An anomaly flag in a finance system that routes a record for action. Often it is switched on by default, and often it is described in release notes that nobody outside the platform team reads.

That path is invisible to every governance artefact a programme maintains. It never passes through procurement, because nothing new was bought. It never reaches a steering committee, because no decision was asked for. It sits outside the scope baseline, because the baseline was set before the release existed. And it is absent from the go-live checklist, because the checklist tests what the programme built, not what the vendor later shipped.

Take a composite example. An organisation implemented a talent acquisition platform three years ago. The programme ran well. The integrator configured screening questions, a knockout rule for work rights and a manual shortlist step, and the steering committee signed off a design in which a recruiter read every application. Two years later a quarterly release added a ranking feature that orders applicants by predicted fit. The platform team enabled it because recruiters asked for it, and it saved them hours a week. Nobody treated it as a decision, because recruiters still clicked the button.

Under APP 1.7 that is almost certainly a computer program doing something substantially and directly related to making a decision about a person, and applicants who are not hired sit outside the employee records exemption. It is also absent from the design document, the steering committee minutes, the closure report and the benefits register. The only record of it is a release note and a configuration change ticket.

Nothing went wrong in that story. Every step was reasonable. That is precisely why no governance artefact caught it.

So when the privacy team asks the business for a list of automated decisions, the business will answer from memory. The list will be short, recent and dominated by anything with “AI” in its name. It will miss the rules engine that has been ranking applicants since 2019.

It is the same pattern Australian audit offices keep finding in technology programmes. The risk is not hidden. It is simply located somewhere no forum is looking.

The four questions a sponsor should be able to answer

A sponsor does not need to become a privacy lawyer. A sponsor does need to be able to answer four questions, with a document rather than an assurance.

Which systems use personal information to make, or substantially shape, a decision about a person? The answer is an inventory, by system and by decision, not a list of AI projects.

Which of those decisions could significantly affect someone’s rights or interests? This is where legal advice belongs, applied to a complete list rather than an impression.

Who switched each capability on, and when? For every automated decision on the list, there should be a record of the configuration decision that enabled it. In most organisations there will not be, which is itself the finding.

Who keeps the list current as vendors release new features? An inventory built for 10 December will be out of date by the next quarterly release. The obligation is continuous, so the ownership has to be.

If the answer to any of the four is “the vendor would know”, the organisation does not yet have an answer.

Who owns the answer

It is tempting to give this to one of three parties. None of them can carry it alone.

The vendor knows what its product can do. It does not know which capabilities the client has enabled, how they are configured, or what decisions they feed. It also has no obligation under APP 1.7. The obligation sits with the organisation that arranged for the computer program to make the decision.

The integrator knows what it configured during implementation. It usually left when the programme closed, and it has no view of what the vendor has released since.

Privacy counsel can interpret the obligation and draft the disclosure. Counsel cannot build the inventory, because the inventory lives in configuration, release notes and business process, not in policy.

The gap between those three is the client-side governance layer. It is the only party that sees the whole estate, holds the business relationships, and has a standing interest in the answer being complete rather than convenient. In a programme, that is the sponsor, the steering committee and whoever provides independent assurance to them. After a programme, it is whoever owns the platform in business-as-usual, which is exactly where accountability for configuration decisions tends to disappear.

What to do in the time remaining

With roughly ten weeks to commencement at the time of writing, the sequence matters more than the checklist.

Start with the inventory, not the policy. Walk each system that touches people, starting with recruitment, contractor management and customer-facing platforms, and record every point where a computer program makes or substantially shapes a decision. Include rules, not just models.

Then take the inventory to legal review. Counsel can now apply the significance test and the exemptions to a complete list, which is the only way the advice is worth having.

Then draft the disclosure. The privacy policy describes kinds of information and kinds of decisions, so the inventory translates into categories, not a catalogue.

Then make it continuous. Add a single line to the change and release process for every people-facing platform: does this release introduce or change an automated decision about a person? If yes, the inventory is updated before the release is approved. That one control does more than any amount of December effort.

And for any programme currently in flight, add the same question to the go-live gate. A system that goes live on 1 December with an undisclosed automated decision is a compliance failure nine days later. The design authority already knows which rules the build contains, so the programme is the cheapest place in the organisation to produce its part of the inventory. Ask for it as a deliverable now, before the integrator’s team rolls off and the knowledge leaves with them.

This article explains an obligation and argues a governance response. It is not legal advice, and any organisation working out its own position should take that advice on its own inventory.

If you want a structured starting point, the AI Readiness assessment walks through where automated capability sits in your estate and who governs it.

Common questions

What is APP 1.7?

APP 1.7 is a new paragraph of Australian Privacy Principle 1, added by the Privacy and Other Legislation Amendment Act 2024. From 10 December 2026 it requires an organisation’s privacy policy to include information about automated decisions, where a computer program uses personal information to make, or do something substantially and directly related to making, a decision that could reasonably be expected to significantly affect an individual’s rights or interests.

When does the automated decision-making disclosure requirement start?

10 December 2026. The infringement and compliance notice powers the OAIC can use for privacy policy failures have been in force since 11 December 2024.

Does APP 1.7 only apply to artificial intelligence?

No. The explanatory memorandum gives “computer program” its ordinary meaning, covering pre-programmed rule-based processes as well as artificial intelligence and machine learning. Long-standing rules engines in recruitment, rostering, finance and customer systems can be in scope.

Does it apply to decisions about employees?

For private-sector employers, the employee records exemption generally takes the handling of current and former employees’ employee records outside the APPs, and the regulator has not yet said how that exemption interacts with the new transparency obligation. The exemption does not cover job applicants who are not hired, volunteers or independent contractors, and it does not apply to Commonwealth agencies. Organisations should take advice on their own position.

What does the privacy policy have to include?

Under APP 1.8, the kinds of personal information used in the operation of the relevant programs, the kinds of decisions made solely by a computer program, and the kinds of decisions where a computer program does something substantially and directly related to making the decision.

What are the penalties for a non-compliant privacy policy?

The OAIC has said entities with non-compliant privacy policies may face compliance and infringement notices and penalties of up to $66,000. It began a compliance sweep of privacy policies in January 2026.