In short
APP 1.7 disclosure starts with operations naming workflows
- APP 1.7 is a privacy policy disclosure rule. It does not require you to publish source code or model weights.
- The obligation is technology-neutral. Rules engines, scoring tools and established software can be in scope alongside generative AI.
- The policy must describe 3 things, set out in full below: the kinds of personal information used and 2 kinds of decisions. Write each so an affected person can understand it.
- Operations has to name the workflows first. Your legal team then confirms the final wording.
Detail
APP 1.7 test and what the privacy policy must describe
Organisations often ask whether they have to disclose their AI. APP 1.7 is not an AI register. It is a privacy policy obligation about some automated decisions, and its test is as follows.
APP 1.7 and APP 1.8 apply from 10 December 2026. They apply where an APP entity has arranged for a computer program to do either of 2 things:
- make a decision
- do a thing that is substantially and directly related to making a decision.
An APP entity is an organisation or agency covered by the Privacy Act 1988. Both of these conditions must also be met:
- 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 operation of the computer program to make the decision or do the thing that is substantially and directly related to making the decision.
The privacy policy must then describe 3 things:
- the kinds of personal information used in the operation of such computer programs
- the kinds of such decisions made solely by the operation of such computer programs
- the kinds of such decisions for which a thing, that is substantially and directly related to making the decision, is done by the operation of such computer programs.
The Office of the Australian Information Commissioner (OAIC) released updated APP 1 guidelines on 30 September 2026. It also released a fact sheet and a flowchart on APP 1.7 to 1.9. Start there, before any vendor presentation.
The guidelines give the term 'computer program' its ordinary meaning. They say a wide range of technologies are likely to fall within it. These include pre-programmed rule-based processes, artificial intelligence, machine learning and commonly used software. A scoring engine that has run the loan book for a decade can matter more than a chatbot launched last month.
The policy describes kinds. You do not list every field and every threshold. A person who might be hired, declined, priced, flagged or restricted should recognise, from the policy, the type of decision that affected them. APP 1.8 lists decisions made solely by a program apart from those where it does a related thing. The OAIC allows the categories to be grouped if the detail is meaningful to a reasonable person. Ask counsel how to present them.
The policy does not have to explain the algorithm. It says that software is part of the decision, what sort of decision it is and what sort of personal information feeds it. The technical documentation, the vendor contract and the override rules stay inside the organisation, where legal, risk and the system owner can use them.
Whether an organisation is in scope is a legal question. Management can start on the operational task now: listing where software already influences an outcome about a person. That includes any use of information about a person who is identified or reasonably identifiable. The privacy policy can describe only the decisions on that list.
Human approval does not automatically take a workflow off the list. The OAIC's guidelines say a program need not make the final decision, or be the only decision-maker, to be captured. Its output can still substantially facilitate, and be directly connected to, a person's decision where they rely on it or are significantly influenced by it.
Commercial impact
Workflow listing is a management task with a fixed date
The date of 10 December 2026 is fixed. After it, a privacy policy that says nothing about automated decisions may be incomplete against an enforceable APP.
Listing the workflows is a management task. Privacy counsel cannot draft kinds of decisions they have not been shown. Operations cannot show them if nobody has mapped the workflows. Vendors may say they are not the APP entity, and the organisation still has a policy to publish.
Starting late compresses legal review and leaves gaps in the list. A board could learn in November that the claims engine, rostering tool and collections scorer were missing from the draft.
Starting late also costs trust. Customers and staff already experience these decisions. If the policy is the first time the organisation admits that software is involved, the disclosure will read as a surprise. An accurate disclosure, published early, shows them the organisation knows where software affects them.
In scope
Decisions the OAIC lists as generally in scope
The OAIC's fact sheet of September 2026 gives a 'non-exhaustive list of decisions that would generally be considered in scope'. They include:
- 'recruitment software used to sort through candidate profiles and make hiring decisions'
- 'computer programs used to approve or reject a loan or credit application'
- 'computer programs involved in assessing an individual's eligibility for insurance'
- 'differential and/or personalised pricing practices used by online retailers in the sale of significant goods'
- 'reports created by artificial intelligence used to rank employee performance, determine promotions, or decide bonus/remuneration outcomes'
- 'facial recognition technology use in a retail store or stadium for watchlist matching'.
Use the list to start your inventory. Then apply the test above to each workflow.
Human review
Human review does not settle scope on its own
The OAIC says a decision may still be in scope where the program's output 'is subject to human review'. It says machine learning or generative AI outputs in significant decisions would 'generally fall within scope' unless 'subject to extensive human oversight and control'.
In its remuneration example, the decision stayed in scope even though a director approved each one in writing. The OAIC says steps such as interrogating outputs and 'documenting the reasoning why it diverged from the recommendation of the computer program' could indicate otherwise.
Treat these as the OAIC's indicators, and ask counsel how they apply to your decisions.
Vendors
Vendor software stays your organisation's question
The OAIC says an organisation can have 'arranged for' a program through 'configuration or customisation of existing off-the-shelf or packaged software'. Relying on 'advisory outputs from a computer program' can count too.
The OAIC 'generally expects that APP 1 transparency obligations typically remain with the APP entity who is using personal information to make a decision'. It says software providers 'should provide clear, high-level information about how their software can be used to make decisions'. Ask your vendors for it now.
Detail
How much detail the privacy policy needs
Organisations 'may group information about the categories of decisions and types of personal information'. The detail 'must be meaningful to a reasonable person'. Where 'health information or biometric templates' are used, the OAIC says 'this should be clear'.
The OAIC asks organisations to 'structure information to enable consumers to request further information'. It also asks them to 'frame information in a way that allows the information and the decision to be reviewed'.
Confidentiality
What can stay confidential
Commercial-in-confidence information is excluded, which the OAIC describes as 'commercially sensitive information' and 'trade secrets'. The exclusion does not cover information whose disclosure 'results in exposure to ridicule, embarrassment or public criticism'.
Significance
Small differences and vulnerable people can be significant
A significant effect 'must be more than trivial' and must be able to 'considerably influence the circumstances or outcomes' for the person, the OAIC says. Vulnerability matters where a decision 'may have a greater impact on persons experiencing vulnerability'.
In the OAIC's pricing example, differential pricing 'is significant regardless of whether the price difference is small or large'.
Refusals
Deciding whom not to serve counts too
In the OAIC's triage example, using a program's rankings to work out 'which individuals not to contact to offer support' would 'still be in scope'. Add the people a program screens out to your mapping session.
Source: OAIC, APP 1.7-1.9 Transparency Obligation Fact Sheet, September 2026.
In practice
Illustrative cases of locating live decision points
Illustrative composite: a mid-market lender assumed APP 1.7 was an AI project. The first review traced the current decisions to a rules engine and a workflow using credit bureau scores. The generative tools the digital team had just bought made no decisions about customers. The privacy policy needed to describe the kinds of decisions from the rules engine and the score workflow, and the chatbot needed no mention.
Illustrative composite: a professional services firm used a vendor screening tool that ranked candidates. Hiring managers usually accepted the first page of the ranking. The legal issue was whether the ranking was a thing substantially and directly related to making the shortlist decision. The practical problem was that nobody could name the ranking's inputs. Both had to be resolved before anyone could write a disclosure sentence.
Illustrative composite: an insurer's claims team described a system recommendation that staff had to follow unless they recorded a reason. Human review does not take a workflow out of scope if the program's output is still a key factor in the decision. The disclosure question followed the workflow, whatever the organisation chart said.
Illustrative composite: a retailer had personalisation and fraud flags in the same vendor platform. One affected what a signed-in customer was offered. The other affected whether an order was held. Those are different kinds of decisions, and merging them into one sentence about AI would have made the disclosure misleading.
What to do
Steps to prepare an APP 1.7 decision map for legal review
- List the software involved in decisions about people before you brief legal, including tools that are not labelled AI.
- For each workflow, record whether a person relies on the program's output, and how.
- Keep decisions made solely by a program apart from those where it does a related thing, because the policy describes them separately.
- Keep the public wording short and the internal map complete. The policy is for readers, and the decision map is for the board and counsel.
- Ask your legal team to confirm whether you are covered and to approve the final privacy policy text.
Last reviewed
APP 1.7 questions on human review and finding decisions
What does APP 1.7 require?
It requires the privacy policy to include the information APP 1.8 lists wherever the test set out above is met. APP 1.9 adds that making a decision includes refusing or failing to make one, so a refusal produced by software counts as a decision. The OAIC published a fact sheet and a flowchart on the obligation on 30 September 2026.
Does APP 1.7 require human review of automated decisions?
APP 1.7 to 1.9 set disclosure rules for privacy policies, and they add no duty to have a person review a decision. Community expectation goes further, according to the OAIC's 2026 Community Attitudes to Privacy Survey. It found 81% of Australians consider a right to have a human review AI decisions essential, up from 73% in 2023.
How do we find the decisions that may be in scope?
Start with each workflow, whatever product runs it. Identify where software uses personal information to affect hiring, eligibility, pricing, claims, account access or service priority. Then check each one against the APP 1.7 test. The APP 1.7 Rapid Review maps those decisions for your privacy and legal teams.
Map the decisions before 10 December 2026.