What must a privacy policy say about automated decisions from 10 December 2026?

Steel rail tracks set into concrete paving cross and split into diverging paths.
Photo: Michael, Pexels
In short From 10 December 2026, an organisation bound by the Privacy Act 1988 (Cth) (‘Privacy Act’), including an overseas company, must say in its privacy policy if it uses personal information in a computer program that makes, or substantially and directly helps make, decisions that could significantly affect individuals. It must list the kinds of information used and of decisions made by software alone or with its substantial and direct help.

Key points

  • Three new clauses in the Australian Privacy Principles of the Privacy Act, APP 1.7–1.9, start on 10 December 2026. They cover decisions from that date that could significantly affect someone, where an organisation bound by the Privacy Act has arranged for software to make the decision or give substantial and direct help in making it.
  • The privacy policy must state the kinds of personal information used, the kinds of decisions made solely by software, and the kinds of decisions in which software does something substantially and directly related to making them.
  • In the guidance of the Office of the Australian Information Commissioner, a person reviewing the software’s output does not by itself take a decision outside the rule. The guidance adds that a business that buys or configures a vendor’s tool can count as having ‘arranged for’ the software to make decisions, which is the first condition of the rule.
  • The duty is limited to the privacy policy, so it gives no right to have a decision explained or reviewed, and the same guidance does not call for an account of how a model works.
  • From 10 December 2026, the Australian Information Commissioner can give an infringement notice, requiring a fixed payment, to an organisation whose privacy policy does not describe automated decisions as required.

If your app approves Australian customers for credit, sets their prices, screens job applicants or blocks accounts with software, your Australian privacy policy has until 10 December 2026 to catch up. From that date, it must say, in plain terms, what kinds of decisions software makes or shapes and what personal information goes into them.1 A policy that stays silent can draw an infringement notice, a demand from the Australian Information Commissioner to pay a fixed amount.2

Two points often surprise businesses. First, the rule reaches decisions that a business’s software makes, or substantially and directly helps make, from 10 December 2026, even if the software and the data are years old.3 Second, the Office of the Australian Information Commissioner (‘OAIC’) interprets the rule broadly in guidance that it published on 30 September 2026. It considers that a decision stays within the rule even if a person reviews the software’s output, a practice often called keeping a human ‘in the loop’. What matters is whether the output is a key factor in what that person decides.4

Who has to add automated decisions to the privacy policy?

From 10 December 2026, the duty can apply to any organisation that the Privacy Act binds. That includes a company based outside Australia that carries on business in Australia, no matter where its servers are.5 Most businesses with annual turnover of $3 million or less are exempt as small businesses. A body corporate, such as a company, is not exempt if related to another body corporate carrying on a business that is not a small business, for example one with turnover above $3 million.6 A separate article explains which companies the Privacy Act binds and who can use the small business exemption.

The rule was inserted by the Privacy and Other Legislation Amendment Act 2024 (Cth) (‘POLA Act’). That Act received Royal Assent on 10 December 2024, and the rule commences two years later, on 10 December 2026.7 It applies to decisions made from that date that meet the three conditions set out below. It does not matter whether the organisation made the arrangement for the program, or the program used the personal information, before or after that date, or when that information was obtained or created.8

An organisation bound by the Privacy Act is caught only if it meets all three conditions in the new APP 1.7.9 The Australian Privacy Principles are the 13 rules in the Privacy Act on handling personal information, and APP 1 is the principle on privacy policies. The table sets out the three conditions and what each means in practice.

Condition What it means in practice
It has arranged for a computer program to make a decision, or to do something substantially and directly related to making one Your own code, a vendor’s platform, a configured off-the-shelf tool, a rules engine (software that applies set rules) or a generative AI model
The decision could reasonably be expected to significantly affect an individual’s rights or interests More than trivial, with the potential to considerably influence the person’s circumstances. Good and bad outcomes both count
Personal information about that individual is used in the program’s operation Name, address, purchase or repayment history, device data, images

The examples in the first two rows are the OAIC’s.10

The OAIC sets out its approach to the three conditions in the APP 1 chapter of its Australian Privacy Principles Guidelines, updated on 30 September 2026 after public consultation. On the same day, it released a fact sheet, APP 1.7–1.9 Transparency Obligation, to help organisations apply the new duty, with practical examples.11

Does a person reviewing the software’s output take a decision outside the rule?

A person reviewing the software’s output does not, as the OAIC understands the rule, take the decision outside it. A program’s part is substantial if its output is a key factor in the person’s decision, and direct if its output has a direct connection with the decision. So, for the OAIC, a recommendation from the software can count as much as a final answer.12 The OAIC takes the position that outputs from machine learning or generative AI used for significant decisions will generally be in scope unless people keep extensive oversight and control of the decision.13

Which decisions ‘significantly affect’ someone?

The Privacy Act gives three examples of decisions that may affect a person’s rights or interests. One is a decision made under a law to grant or refuse a benefit, such as a government payment. The others are decisions that affect a person’s rights under a contract, and decisions that affect their access to a significant service or support. The effect must still be significant. Refusing or failing to make a decision also counts as making one, and refusing or failing to do something counts as doing it.14

The OAIC’s fact sheet goes further, with its own list of decisions that would generally be in scope. The list includes:15

  • approving or rejecting a loan or credit application;
  • personalised or differential pricing, where an online retailer charges different people different prices for significant goods;
  • deciding whether a person is eligible for insurance;
  • recruitment software that sorts candidates, and AI reports used to rank staff or set bonuses;
  • programmatic advertising, where software automatically chooses which advertisements a person sees, tied to the sale of significant goods or services.

Even small amounts can make a decision significant. The OAIC illustrates the point with an online grocery platform that prices essentials such as baby formula by postcode. Each price difference is under $4 an item, yet one shopper paid $132 more for baby formula over a year. The OAIC regards the pricing as significant because small differences add up over time and the goods include food and medication.16

We use a vendor’s software. Is the disclosure our job or theirs?

The OAIC sees the disclosure as usually the job of the business that uses a vendor’s software to make decisions, not of the vendor. Buying a third-party program, configuring an off-the-shelf product or relying on a program’s recommendations can each mean that the business has ‘arranged for’ the program, which is the first condition.17 The OAIC encourages vendors to give their customers clear, high-level information about how their software can be used to make decisions.18

In practice, the duty also stays with the Australian business when a parent or related company abroad runs the software that scores customers or sets prices. The Australian business then needs enough information from that company to keep its own policy accurate and up to date.19

What exactly must the policy say?

The policy must describe three things, each by kind rather than case by case:

  1. the kinds of personal information used in the operation of those programs;
  2. the kinds of decisions made solely by the programs;
  3. the kinds of decisions for which a program does something substantially and directly related to making the decision, even though the program does not make the decision alone.20

These three things sit alongside what APP 1.4 already requires in every privacy policy, such as the kinds of information collected and the purposes for which it is collected. The policy must also say whether the information is likely to be disclosed to recipients overseas and, where practicable, in which countries.21 The new duty concerns only what the privacy policy says. It does not give an individual a right to ask the business to explain a particular decision, or to have a person review it.22

The OAIC’s guidance does not ask for technical detail of how a model works. It allows decisions and kinds of information to be grouped together, as long as the description still means something to a reasonable person. It also emphasises that any use of sensitive information should be made clear. Examples are health information and biometric templates, the stored digital records of a person’s face or fingerprint used to recognise them.23

The OAIC concludes, drawing on the government’s notes to Parliament on the bill that became the POLA Act, that commercially sensitive information and trade secrets need not be disclosed. It keeps that exclusion narrow: the exclusion covers only information whose release could harm the business’s commercial interests.24 An OAIC example shows where the line falls: a lender need not disclose how its fraud tool weights each piece of data, but must disclose that the tool uses customers’ personal information.25

What happens if the privacy policy is not updated by 10 December 2026?

From that date, APP 1.7 joins the list of privacy policy duties in s 13K(1), beside the existing duties to have a policy and to include the APP 1.4 contents.26 A breach of any of them can lead to an infringement notice, which the Commissioner, or a senior executive on the OAIC’s staff, can issue without going to court.27 The notice states 200 penalty units ($72,800) for each alleged breach by a listed corporation, and 60 penalty units ($21,840) for any other company.28

If a notice is not paid, the Commissioner can ask a court for a civil penalty. For each breach by a company, the Federal Court or the Federal Circuit and Family Court of Australia can order up to 1,000 penalty units ($364,000).29 A breach of APP 1.7 is also an ‘interference with privacy’. The Privacy Act notes that the same conduct may also breach the provisions that carry higher maximum penalties for an interference with privacy.30 A separate article explains those higher maximums and the factors that make a breach serious.

In practice, a gap in a privacy policy is easy to find, because the policy must be available free of charge and usually sits on the company’s website.31

Worked example: what must a Shenzhen marketplace’s privacy policy say?

The example below is hypothetical. Its systems are tested in turn against the conditions and the disclosure duty set out earlier.

Take a Shenzhen-headquartered online marketplace that sells household goods to Australians through its app. Its software sets prices by postcode and purchase history, and a credit model uses each shopper’s purchase and repayment history to decide whether the shopper may pay in four instalments. A fraud tool scores accounts on their payment, device and location data, suspends some on its own and sends others to a review team in Shenzhen, who usually follow its score. Assume also that the marketplace carries on business in Australia, so that the Privacy Act binds it, and that it set up its pricing software, credit model and fraud tool in 2025.

First, the marketplace has arranged for computer programs that use shoppers’ personal information to make each kind of decision, so the first and third conditions are met. It does not matter that the marketplace set the programs up before 10 December 2026.32 The suspensions decided by staff meet the first condition too, because the fraud score is a key factor in the review team’s decision, which is enough by the OAIC’s test.33 What remains is whether each kind of decision could significantly affect shoppers.

Second, deciding whether a shopper may pay in instalments is, in substance, a credit decision, which the OAIC lists as generally in scope. A suspension affects the shopper’s rights under their contract, one of the Privacy Act’s own examples, and in practice is likely to be significant.34 Postcode pricing of household goods is less clear: whether it is significant depends on the goods, the size of the price differences and the shoppers affected, and the OAIC’s grocery example concerned food and medication. Because the answer is uncertain, the OAIC’s fact sheet tells an organisation in doubt to include the information.35

Third, from 10 December 2026, the marketplace’s privacy policy would need to say, in substance:36

  • Information used: postcode, purchase and payment history, repayment record, device and location data.
  • Decisions made solely by software: approving or declining instalment payment; automatic account suspension for suspected fraud; personalised prices.
  • Decisions where software plays a substantial and direct part: suspensions decided by staff after a fraud score.

The policy need not reveal the fraud tool’s weightings if they are commercially sensitive or a trade secret in the OAIC’s narrow sense, but it must still say that the tool uses shoppers’ personal information.37 Finally, if the policy leaves these decisions out, the marketplace risks an infringement notice for 60 penalty units ($21,840) for each alleged breach.38

What steps should a business take before 10 December 2026?

For a business that uses software to make or shape decisions about people in Australia, as the marketplace in the example does, the work before 10 December 2026 comes down to five steps.

  1. List every program that makes, or substantially and directly helps to make, decisions about individuals, including vendors’ tools and group systems overseas. As explained above, buying or configuring a tool can count as arranging for it.39
  2. Check each program against the three conditions in APP 1.7. Where the answer is doubtful, include it, in line with the OAIC’s guidance.40
  3. Ask vendors and group companies for the information needed to describe their programs.41 In practice, it is easier to ask now than in December 2026.
  4. Add a section to the privacy policy that states the three things explained above, and make any use of sensitive information clear.42
  5. Publish the updated policy before 10 December 2026, and keep it up to date as the programs change.43

What is not settled (as at 4 November 2026)

  • Neither the OAIC nor a court has yet applied APP 1.7–1.9 in a formal decision, because the rule is not in force. Until one does, the OAIC’s guidance, explained above, is the main guide.
  • The next privacy bill could change other parts of the same policy. The exposure draft of the Privacy Amendment (Personal Data Protection) Bill 2026 leaves APP 1.7–1.9 and their start date untouched, but it would change other rules that the policy must reflect, such as the rules on direct marketing.44
  • How often the OAIC will issue infringement notices for APP 1.7, and in which sectors, is not yet known.

Questions

Do we have to explain how our algorithm works?
A privacy policy need not explain an algorithm in technical detail. It must describe the kinds of personal information and decisions involved in a way that a reasonable person can understand.
Does the rule give Australian customers a right to an explanation or a human review?
As written, APP 1.7–1.9 give no right to an explanation or a human review. They set what the privacy policy must contain. They do not create a right to have a particular decision explained or reviewed.
Our systems were built years ago. Are they covered?
Older systems are covered for decisions they make, or give substantial and direct help in making, from 10 December 2026, if the decisions could significantly affect people and use their personal information. It does not matter when the business set up the software, when the software first used the personal information, or when that information was obtained or created.
Is the 10 December 2026 date likely to move?
As at 4 November 2026, the start date has not moved. From 10 December 2026, an organisation bound by the Privacy Act that uses software in decisions covered by APP 1.7–1.9 must describe those automated decisions in its privacy policy. The draft of the next privacy bill does not change APP 1.7–1.9 or their start date.
Anna Zheng, principal of SQ Legal
About the author

Anna Zheng is an Australian lawyer and the principal of SQ Legal. She advises Australian and international companies on data and privacy. Her data and privacy work includes reviewing what an Australian business automates and rewriting its privacy policy and collection notices to match.

Contact Anna at [email protected]Email[email protected]Open in your default email appGmailOutlook (personal)Outlook (work or school)Or copy the address:[email protected].

This article is general information about Australian law as at the date shown above. It is not legal advice and does not take account of your circumstances. Reading it does not create a lawyer–client relationship with SQ Legal. Obtain advice on your own matter before acting. See the terms of use.

More on data and privacy, or contact SQ Legal about a matter.

Cite this article

Anna Zheng, ‘What must a privacy policy say about automated decisions from 10 December 2026?’, SQ Legal (Web Page, 16 November 2026) <https://sqlegal.com.au/insights/automated-decisions-privacy-policy/>.

You may quote short extracts with attribution and a link to this page. To republish the article in full, write to [email protected]Email[email protected]Open in your default email appGmailOutlook (personal)Outlook (work or school)Or copy the address:[email protected].

Footnotes

  1. Privacy and Other Legislation Amendment Act 2024 (Cth) s 2(1) item 7, sch 1 item 88 (‘POLA Act’), inserting Privacy Act 1988 (Cth) sch 1 cls 1.7–1.9 (‘Privacy Act’). ↩

  2. POLA Act (n 1) sch 1 item 87, inserting Privacy Act (n 1) s 13K(1)(b)(iia); Privacy Act (n 1) s 80UB(1)(a), (2). ↩

  3. POLA Act (n 1) sch 1 item 89. ↩

  4. Office of the Australian Information Commissioner, ‘Chapter 1: APP 1 Open and Transparent Management of Personal Information’ in Australian Privacy Principles Guidelines (Guidelines, 30 September 2026) [1.41], [1.50] (‘APP 1 Guidelines’); Office of the Australian Information Commissioner, APP 1.7–1.9 Transparency Obligation (Fact Sheet, September 2026) 2 (‘Fact Sheet’). ↩

  5. Section 5B(3) sets no condition about where personal information is collected or held: Privacy Act (n 1) s 5B(1A), (3). ↩

  6. Privacy Act (n 1) ss 6C(1) (definition of ‘organisation’), 6D(1), (9). ↩

  7. POLA Act (n 1) s 2(1) item 7. ↩

  8. Ibid sch 1 item 89. ↩

  9. Ibid sch 1 item 88, inserting Privacy Act (n 1) sch 1 cls 1.7, 1.9. ↩

  10. Fact Sheet (n 4) 1–2; APP 1 Guidelines (n 4) [1.47], [1.57]. ↩

  11. Office of the Australian Information Commissioner, ‘New Resources on Transparency for Use of AI and Automated Decision-Making’ (Media Release, 30 September 2026); APP 1 Guidelines (n 4); Fact Sheet (n 4) 1. ↩

  12. APP 1 Guidelines (n 4) [1.41], [1.50], [1.52]; Fact Sheet (n 4) 2. ↩

  13. Fact Sheet (n 4) 2. ↩

  14. POLA Act (n 1) sch 1 item 88, inserting Privacy Act (n 1) sch 1 cls 1.7(b), 1.9(a)–(d). ↩

  15. Fact Sheet (n 4) 5. ↩

  16. Ibid 7–8. ↩

  17. Fact Sheet (n 4) 2–3; APP 1 Guidelines (n 4) [1.44]–[1.45]. ↩

  18. Fact Sheet (n 4) 3. ↩

  19. The organisation that arranged for the program must keep its own policy up to date: Privacy Act (n 1) sch 1 cl 1.3; Fact Sheet (n 4) 3. ↩

  20. POLA Act (n 1) sch 1 item 88, inserting Privacy Act (n 1) sch 1 cl 1.8. ↩

  21. Privacy Act (n 1) sch 1 cl 1.4(a), (c), (f)–(g). ↩

  22. POLA Act (n 1) sch 1 item 88, inserting Privacy Act (n 1) sch 1 cls 1.7–1.9. ↩

  23. Fact Sheet (n 4) 4; APP 1 Guidelines (n 4) [1.67], [1.70]–[1.71]. ↩

  24. APP 1 Guidelines (n 4) [1.67]; Fact Sheet (n 4) 3. ↩

  25. Fact Sheet (n 4) 9–10. ↩

  26. POLA Act (n 1) sch 1 item 87, inserting Privacy Act (n 1) s 13K(1)(b)(iia); Privacy Act (n 1) s 13K(1)(b)(i)–(ii). ↩

  27. Privacy Act (n 1) s 80UB(1)(a), (2); Regulatory Powers (Standard Provisions) Act 2014 (Cth) s 103(1) (‘Regulatory Powers Act’). ↩

  28. For a company that is not listed, the amount is the lesser of one-fifth of the court maximum of 1,000 penalty units and 60 penalty units: Privacy Act (n 1) ss 13K(4), 80UB(1A)–(1B); Regulatory Powers Act (n 27) ss 82(5)(a), 104(2)–(3); Crimes Act 1914 (Cth) s 4AA(1), (1A), (3); Crimes (Amount of a Penalty Unit) Instrument 2026 (Cth) s 5. ↩

  29. Regulatory Powers Act (n 27) ss 82(1), (5)(a), 104(1)(k)(iii), 108(b)(i); Privacy Act (n 1) ss 13K(4), 80U(1)–(3). ↩

  30. Privacy Act (n 1) ss 13(1)(a), 13G(1), (3), 13H(1), (3), and the note to s 13K(1). ↩

  31. Ibid sch 1 cl 1.5, including the note to cl 1.5. ↩

  32. POLA Act (n 1) sch 1 item 88, inserting Privacy Act (n 1) sch 1 cl 1.7(a), (c); POLA Act (n 1) sch 1 item 89. ↩

  33. APP 1 Guidelines (n 4) [1.50], [1.52]. ↩

  34. Fact Sheet (n 4) 5; POLA Act (n 1) sch 1 item 88, inserting Privacy Act (n 1) sch 1 cls 1.7(b), 1.9(d)(ii). ↩

  35. Fact Sheet (n 4) 1, 7–8. ↩

  36. POLA Act (n 1) sch 1 item 88, inserting Privacy Act (n 1) sch 1 cl 1.8. ↩

  37. APP 1 Guidelines (n 4) [1.67]; Fact Sheet (n 4) 9–10. ↩

  38. POLA Act (n 1) sch 1 item 87; Privacy Act (n 1) ss 13K(4), 80UB(1)(a), (1A)–(1B); Regulatory Powers Act (n 27) ss 82(5)(a), 104(2)–(3); Crimes Act 1914 (Cth) s 4AA(1), (1A), (3); Crimes (Amount of a Penalty Unit) Instrument 2026 (Cth) s 5. ↩

  39. APP 1 Guidelines (n 4) [1.44]–[1.45]; Fact Sheet (n 4) 2. ↩

  40. Fact Sheet (n 4) 1–2. ↩

  41. Ibid 3. ↩

  42. POLA Act (n 1) sch 1 item 88, inserting Privacy Act (n 1) sch 1 cl 1.8; APP 1 Guidelines (n 4) [1.71]. ↩

  43. Privacy Act (n 1) sch 1 cl 1.3. ↩

  44. The exposure draft amends s 13K and replaces APP 7, but does not amend sch 1 cls 1.7–1.9 of the Privacy Act or the POLA Act: Exposure Draft, Privacy Amendment (Personal Data Protection) Bill 2026 (Cth) sch 2 items 14–16, 26–27; ‘Privacy Reform: Consultation on Exposure Draft Legislation’, Attorney-General’s Department (Web Page, 31 August 2026) <https://consultations.ag.gov.au/rights-and-protections/privacy-reform/>. ↩

Sources

  1. Privacy and Other Legislation Amendment Act 2024 (Cth)
  2. Privacy Act 1988 (Cth)
  3. Regulatory Powers (Standard Provisions) Act 2014 (Cth)
  4. Crimes Act 1914 (Cth)
  5. Crimes (Amount of a Penalty Unit) Instrument 2026 (Cth)
  6. Exposure Draft, Privacy Amendment (Personal Data Protection) Bill 2026 (Cth)
  7. Office of the Australian Information Commissioner, ‘Chapter 1: APP 1 Open and Transparent Management of Personal Information’ in Australian Privacy Principles Guidelines (Guidelines, 30 September 2026)
  8. Office of the Australian Information Commissioner, APP 1.7–1.9 Transparency Obligation (Fact Sheet, September 2026)
  9. Office of the Australian Information Commissioner, ‘New Resources on Transparency for Use of AI and Automated Decision-Making’ (Media Release, 30 September 2026)
  10. ‘Privacy Reform: Consultation on Exposure Draft Legislation’, Attorney-General’s Department (Web Page, 31 August 2026)