From 10 December 2026, any Australian organisation covered by the Privacy Act that uses a computer program to make, or substantially help make, decisions that significantly affect people must say so in its privacy policy. The policy has to list the kinds of personal information those programs use and the kinds of decisions they make or support, and the rule applies to systems already running today, not just new ones.
The change comes from Part 15 of the Privacy and Other Legislation Amendment Act 2024, which added three new subclauses, 1.7 to 1.9, to Australian Privacy Principle 1. It is a disclosure rule, not a ban, but you can’t disclose what you haven’t found. For most mid-market organisations, the real work is finding every rules engine, scoring model, AI agent and vendor feature that sits inside a decision about a customer, employee or applicant.
This guide covers what the law says, which systems it catches, what your privacy policy needs to say, and a checklist to finish before 10 December. It is general information, not legal advice. Have your lawyers review your final privacy policy wording.
What exactly does the new rule require?
APP 1.7 says your privacy policy must include the information in APP 1.8 when three conditions are all true:
- you have arranged for a computer program to make a decision, or to do a thing that is “substantially and directly related to making” a decision
- the decision could reasonably be expected to significantly affect an individual’s rights or interests
- personal information about that individual is used when the program runs
If all three apply, APP 1.8 says the policy must set out:
- the kinds of personal information used in the operation of those programs
- the kinds of decisions made solely by those programs
- the kinds of decisions where a program does something substantially and directly related to the decision, even though a person makes the final call
APP 1.9 widens the rule in two ways. Refusing or failing to make a decision counts as making one. And an effect counts whether it helps or harms the person. The Act gives three examples of decisions that can affect rights or interests: granting or refusing a benefit under legislation, decisions that affect a person’s rights under a contract or arrangement, and decisions that affect access to a significant service or support.
The obligation commences on 10 December 2026, 24 months after Royal Assent. Item 89 of the amending Act makes it apply to any decision made after that date. That holds even if the system was set up earlier and the personal information was collected years ago.
Does it apply to AI agents, or to simple rules as well?
It applies to both. The OAIC’s automated decision-making issues paper, released on 18 May 2026, reads “computer program” broadly. Norton Rose Fulbright’s summary of the paper notes that it covers pre-programmed rule-based processes, machine learning and generative AI tools. A spreadsheet formula can count. An AI agent built on a large language model can count. The test is the decision and its effect on a person, not how sophisticated the technology is.
Two phrases do most of the work:
- “Arranged for.” The obligation sits with the organisation that arranged for the program to be used, not with whoever built or hosts it. Decisioning features in a vendor’s SaaS platform are still your disclosure.
- “Substantially and directly related.” MinterEllison’s analysis describes the OAIC’s position this way: “substantially” means the output is “a key factor in facilitating the human’s decision making”. “Directly” means there is “a direct connection with the making of the decision”. A human approver does not take a process out of scope if that person mostly acts on the program’s score or recommendation.
That last point matters most for agentic AI. Many teams treat “human in the loop” as a compliance shield. Under APP 1.7, an agent that drafts the recommendation a person rubber-stamps is still doing a thing substantially and directly related to the decision. The disclosure still applies. Keeping a person in the loop changes which APP 1.8 category the decision falls into. It doesn’t remove the duty to disclose.
Which of our systems are likely to be caught?
Start with decisions, not systems. Ask where your organisation says yes or no to a person, sets their price or terms, or decides how quickly they get help. Then trace which software touches each of those decisions. The table below is our working triage for a typical Australian mid-market stack. Test your own facts against it with legal advice.
| Decision | Where it often lives | Likely in scope? | Why |
|---|---|---|---|
| Approving or declining credit, payment terms or a refund | ERP, billing platform, payment gateway rules | Likely | Affects rights under a contract or arrangement |
| Shortlisting or rejecting job applicants | Applicant tracking system screening and ranking | Likely | Shapes access to employment; named in the OAIC consultation examples |
| Personalised pricing or eligibility for an offer | CDP, commerce platform, loyalty engine | Possibly | Depends on how significant the price or offer difference is |
| Routing, prioritising or closing a complaint or service request | Service desk automation, triage agents | Possibly | Higher risk where the service is significant, such as health, utilities or financial hardship |
| Flagging fraud or suspending an account | Fraud rules, identity verification services | Likely | Can cut off access to a service |
| Ranking sales leads for follow-up | CRM lead scoring | Usually not, but check | A sales call order rarely affects a person’s rights significantly |
| Choosing which marketing email someone receives | Marketing automation journeys | Usually not | Unlikely to be significant on its own; check APP 7 direct-marketing rules separately |
CRM lead scoring is a useful boundary case. Salesforce’s documentation for Einstein Lead Scoring explains that it analyses historical lead fields, including custom fields, to predict which leads to prioritise. Ranking B2B leads for a sales team is unlikely to significantly affect anyone. The same scoring pattern used to decide who gets a payment plan or a service upgrade is a different matter. Mapping which Salesforce and CRM automations feed which decisions is usually the first job, because those platforms make scoring and routing easy to switch on without a privacy review.
What should the privacy policy actually say?
APP 1.8 asks for kinds of information and kinds of decisions. It doesn’t ask for algorithms, model weights or source code. The OAIC has said it will publish final guidance before commencement. Check oaic.gov.au for the published version before you finalise wording. In the meantime, a clear disclosure is grouped by decision and written in plain language. For example:
Decisions made solely by computer programs. We use automated rules to approve or decline refund requests under $200. These rules use your order history, payment method and previous refund requests.
Decisions where computer programs play a substantial role. When you apply for a role with us, software ranks applications against the selection criteria using your CV, answers to screening questions and work-rights status. A member of our recruitment team reviews the ranking and makes the decision.
Treat the example above as a pattern to adapt, not text to copy. The point is specificity: name the decision, the category of information and the role the software plays. A single line saying “we may use automated tools” will not meet APP 1.8.
What happens if we get it wrong?
The amending Act added APP 1.7 to section 13K of the Privacy Act. That section lists the APP obligations the Information Commissioner can enforce through infringement notices and compliance notices, without first going to court. For a listed corporation, the Act sets the amount in an infringement notice at 200 penalty units for each alleged contravention. An inaccurate or missing ADM disclosure is now an administrative breach the regulator can act on quickly. A larger exposure follows if the disclosure exercise uncovers a deeper problem, such as personal information used for a purpose people would not expect. That falls under APP 6 and the broader interference-with-privacy penalties.
The reputational risk is harder to price. Your privacy policy becomes a public list of where software makes decisions about people. A precise disclosure reads as good governance; a vague one invites questions.
What should we do before 10 December 2026?
Eight weeks is enough time for most mid-market organisations if the work starts now. Use this checklist:
- Name an owner. Give one person accountability for the ADM inventory and the policy update. This matches the first practice in the federal Guidance for AI Adoption, “decide who is accountable”.
- List significant decisions. Ask each business unit which decisions about customers, staff, applicants or suppliers’ staff could affect their rights, contracts or access to services.
- Trace the software behind each decision. Include ERP and CRM workflow rules, scoring models, AI agents, vendor decisioning features and the integrations that pass data between them.
- Classify each one. Decide whether it is made solely by a program, substantially supported by one, or out of scope. Record your reasoning.
- Record the personal information used. Use your integration and data maps rather than vendor marketing material.
- Check your vendor contracts. Confirm vendors will tell you when they add or change decisioning features, because you carry the disclosure.
- Draft the APP 1.8 disclosure in plain language, grouped by decision type, and have it reviewed by legal counsel.
- Put a change trigger in your delivery process. Any new automation, agent or scoring model that touches a significant decision should prompt a privacy policy review before go-live.
- Keep a register. The Department of Industry provides an AI register template with the guidance. Extend it to cover rule-based automation as well as AI.
- Publish before 10 December 2026, then review the disclosure at least every six months.
Why is this an integration problem as much as a legal one?
The decisions APP 1.7 cares about rarely sit in one system. A refund decision might draw order history from the commerce platform, risk flags from a fraud service and customer status from the CRM, then post the outcome to the ERP. No single system owner can see the whole decision. That is why so many organisations will struggle to answer the first question their lawyers ask: which programs, using which personal information?
Organisations with documented integrations will finish this quickly. Those without them will find that the ADM inventory becomes an integration audit. That is not wasted effort. The same map supports Privacy Act tranche 2 reforms, AI governance and any future platform consolidation. We wrote about the delivery side of this in agentic AI in systems integration. Our broader view of where agents add value is in Agentic AI: unlocking scalable intelligence. For an example of agents operating on product data rather than people, see the role of agentic AI in PIM integration.
Workflow automation is where this shows up first. In our Salesforce CRM implementation for a Hong Kong law firm, we automated document generation, approvals and task assignment for a 100-person firm. Document processing time fell by 40%. Automated approval steps like these are exactly where an ADM review should start. For each one, ask whether it sits inside a decision about a person; if it does, the answer belongs in the privacy policy as well as the solution design.
Frequently asked questions
Does the rule apply to small businesses?
It applies to APP entities. Most businesses with annual turnover of $3 million or less are exempt from the Privacy Act, with exceptions such as health service providers and businesses that trade in personal information. Check whether your organisation is covered before deciding the rule doesn’t apply.
Does a human reviewer take a decision out of scope?
Not if the program’s output is a key factor in that person’s decision. That decision then falls under the “substantially and directly related” category in APP 1.8(c) and still has to be disclosed.
We use a vendor’s AI feature. Is that their disclosure or ours?
Yours, if you arranged for the program to be used in your decisions. Ask vendors for documentation of what their decisioning features do and what data they use.
Where to start
If you are unsure which of your platforms make or shape decisions about people, start with a short inventory workshop. Bring your service, CRM, ERP and HR system owners into one room for two hours and walk each significant decision from end to end. Our agentic AI and data strategy teams can run that session with you and produce the register your privacy lawyers will need. Contact us to book a time before the December deadline.