Skip to content
Briefing 0311 minute readRisk and governance

Automated decisions: the 10 December privacy deadline

From 10 December 2026, Australian privacy policies must say where computer programs make or shape decisions about people. A person signing off does not take a process out of scope. Here is what to map and what to write.

Tom BuckleySpurbuck Advisory
79 in 100 Australians want to be told when AI is usedOAIC community attitudes survey, 2026

The short version

  • From 10 December 2026, an organisation covered by Australia's Privacy Act must say in its privacy policy when a computer program makes, or substantially helps make, decisions that could significantly affect people, according to the OAIC's updated guidelines 1.
  • A person signing off does not take a process out of scope: the OAIC says a program is captured where the human decision-maker relies on or is significantly influenced by its output 1.
  • The rule is not only about AI. The OAIC reads the term computer program broadly, to include rule-based processes, and the law firm Allens reads the guidance as reaching everyday software such as spreadsheets 1, 5.
  • In the OAIC's 2026 community attitudes survey, 79% of Australians said they should be informed when AI is used, and only 25% found automated eligibility or risk decisions acceptable 3.
  • The work is an inventory before it is a rewrite: list where programs feed decisions about people, test each one against three questions, then describe the kinds of decisions and information in plain English.
10 Dec2026: the day privacy policies must start describing automated decisions 1OAIC, APP guidelines
79%of Australians want to be informed when AI is being used, up from 71% in 2023 3OAIC survey, 2026
81%want the right to have a human review decisions, up from 73% in 2023 3OAIC survey, 2026
25%find automated eligibility or risk-based decisions acceptable 3OAIC survey, 2026

What changes on 10 December

In late 2024 Parliament passed the Privacy and Other Legislation Amendment Act. One Part of it deals with automated decisions and privacy policies 9, and it takes effect on 10 December 2026 1. From that day, an organisation covered by the Privacy Act has to add something to its privacy policy if computer programs play a part in its decisions about people.

The Office of the Australian Information Commissioner (OAIC) published its guidance on 30 September: a fact sheet, a flowchart and an updated chapter of its guidelines on the Australian Privacy Principles 2. The test in those guidelines has three parts, and all three have to be met 1.

Three questions decide whether a process is in scopeA process is covered only if the answer to all three is yes. Paraphrased from the OAIC's guidelines 1.
  1. Does a program make the decision, or substantially shape it?The program

    You have arranged for a computer program to make a decision, or to do something substantially and directly related to making it.

  2. Could the decision significantly affect a person?The effect

    It could reasonably be expected to significantly affect an individual's rights or interests. Helpful decisions count as well as harmful ones.

  3. Does the program use that person's information?The data

    Personal information about the individual is used when the program runs.

Where all three are met, the privacy policy must describe three things: the kinds of personal information the programs use, the kinds of decisions made solely by a program, and the kinds of decisions where a program does something substantially and directly related to the decision 1, 2.

Who is covered? Broadly, organisations with annual turnover above $3 million, plus some smaller ones such as private health providers and businesses that buy or sell personal information 8. Most mid-sized Australian organisations are in.

A person signing off does not take you out of scope

This is the part most organisations will get wrong. It is natural to read an automated decision as a machine acting alone, and to conclude that any process with a person in it is exempt. The OAIC says otherwise. A program is captured where the human decision-maker "relies on or is significantly influenced by" what it produced 1.

The guidelines list what to weigh: how much the person relies on the output, whether they can and do override it, whether the output is advice or the answer, how well it can be explained, and how deeply the program is built into the workflow 1. One of the OAIC's examples is a program that ranks scholarship applications for a panel that is free to ignore the ranking. It is still in scope 1.

Two more points widen the net. First, the term computer program takes its ordinary meaning. It includes rule-based processes as well as machine learning and generative AI 1, and the law firm Allens reads the guidance as reaching everyday software, spreadsheets included 5. A scoring formula that has sat in a workbook for ten years counts if it meets the three tests.

Second, the duty belongs to whoever arranged for the program to be used, not whoever wrote it 1. Buying the tool from a vendor does not move the duty to the vendor 5. Informal tools don't escape either. In another OAIC example, an employer that lets staff use an AI chat tool to draft performance assessments for promotion decisions has arranged for that program 1. Each of these still has to pass the other two questions before it needs disclosing.

In the frame, on the OAIC's examples

  • Vendor software that approves or declines refunds
  • A program that ranks scholarship applicants for a panel
  • An AI chat tool staff use to draft performance assessments for promotions

Outside it

  • Word processing and other tools that feed no decision
  • A decision whose effect on the person is trivial
  • A program that uses no personal information

What counts as a significant effect

The second question is where judgement comes in. The OAIC says the effect has to be more than trivial, with the potential to considerably influence a person's circumstances, and that the person's situation matters: the same decision can be minor for one customer and serious for a child or a vulnerable person 1.

The guidelines give a list, which they say is not exhaustive, of the interests they have in mind: access to healthcare and aged care, financial assistance, education, essential banking and credit, telecommunications, utilities, employment and housing 1. In some cases targeted pricing can qualify too 1.

Our reading is that, for a mid-sized business, this points at a familiar set of processes: credit and payment terms, tenancy and loan applications, insurance claims, recruitment screening, performance ratings and account suspensions. Look there first.

What Australians want before AI is used to decideShare of Australian adults who say each condition should be in place before organisations use AI to make decisions. OAIC Australian Community Attitudes to Privacy Survey 2026, 1,504 respondents 3.

The law asks for less than your customers do

The new rule is a disclosure rule. Allens describes it as a transparency measure only: it gives people no right to contest a decision or to ask for information about one, and it does not require you to notify anyone individually 5. Nor do you have to publish how a model works: the OAIC accepts that commercial-in-confidence detail can stay out, though it treats that as a narrow exception 1.

Australians expect more than that. In the OAIC's 2026 survey of 1,504 adults, 96% said some conditions should be in place before AI is used to make decisions. Four in five (81%) wanted the right to have a person review a decision, up from 73% in 2023, and 79% wanted to be told when AI is being used, up from 71% 3. Asked about specific uses, 64% found AI acceptable for detecting fraud. Only 25% said the same of automated eligibility or risk decisions such as loan approvals 3.

So the paragraph you add is the legal minimum, and it is also the first time many customers will read, in your own words, that software helps decide things about them. A policy that says a program scores applications and a named team reviews every refusal reads very differently from one that says automated tools may be used.

The regulator is already reading privacy policies

The OAIC began 2026 with its first compliance sweep, reviewing the privacy policies of about 60 businesses in six sectors against the existing rules on what a policy must contain. It said then that a non-compliant policy could bring compliance and infringement notices and penalties of up to $66,000, using powers Parliament added in 2024 4.

That sweep was about what privacy policies were already supposed to contain, not the new automated-decision paragraphs. The lesson carries over: the regulator now reads policies without waiting for a complaint, and from 10 December there is one more thing a policy has to say.

New Zealand: no matching rule, the same direction

New Zealand's Privacy Commissioner starts from the position that the Privacy Act 2020 applies to everyone using AI tools in New Zealand, and recommends a privacy impact assessment before an organisation starts using one 7. As we read it, the Commissioner's AI guidance sets no disclosure rule of the Australian kind.

What New Zealand did add this year is a notification duty. Since 1 May 2026, under a new information privacy principle known as IPP 3A, an agency that collects personal information from someone other than the person concerned must take reasonable steps to make sure that person knows 6. An agent that enriches a customer record from a third-party source runs straight into it. For organisations on both sides of the Tasman, one inventory of where programs feed decisions about people serves both.

Five steps before 10 DecemberIn order. The first two take the time; the writing is quick once they are done.
PilotProduction12345
  1. 1List every program that feeds a decision about a person. Customers, applicants and staff. Include vendor tools, the scoring formula in a spreadsheet and any AI chat tool people use for assessments 1, 5.
  2. 2Test each one against the three questions. Does a program make or substantially shape the decision, could the effect on a person be significant, and does it use their personal information? Write down the answer and the reason 1.
  3. 3Check who decides, in the contract and in practice. Where a vendor's software is involved, the duty stays with you. The OAIC expects you to keep oversight of third-party programs and to be clear about who holds the final say 1.
  4. 4Write the disclosure in plain English. The kinds of personal information used, the kinds of decisions a program makes alone, and the kinds where a program substantially shapes a person's decision. Name sensitive information such as health or biometric data plainly 1.
  5. 5Make it part of go-live. Test every new automation or agent against the three questions before it starts, so the policy stays true.

What this means if you are building AI agents

Much of the work on AI agents this year has been about keeping a person in charge: the agent prepares, a person approves. That remains the right design. It is not an exemption. If an agent's draft or score is a key factor in a decision that matters to someone, then on the OAIC's reading the process belongs in your privacy policy whether or not a person presses the button 1.

The better news is that a well-built agent already carries most of what the rule needs. An agent with a written scope has a list of what it may read, what it may decide alone and what waits for approval, which is the raw material for the disclosure. If no such record exists for the automations you already run, the gap is worth closing for its own sake.

This briefing is general information, not legal advice. Allens notes that the guidance leaves significant room for interpretation, particularly on what counts as a significant effect 5, so have your privacy adviser confirm how the three questions apply to your own processes.

Questions for your board before 10 December

If nobody can answer the first one, start there.

  1. Do we have a list of every place a computer program makes or shapes a decision about a customer, an applicant or an employee?
  2. Which of those could significantly affect someone, and who made that call?
  3. Where a person signs off, how often do they overrule the program?
  4. Which of these programs came from a vendor, and does the contract say who is making the decision?
  5. Who owns the privacy policy update, and when will the board see the wording?
  6. What stops a new automation going live without being checked against this rule?

Need the inventory as well as the wording?

Our AI Opportunity Assessment includes a controls plan for each automation: who is accountable, what it may do alone and what needs approval. It is the record a disclosure like this is written from.

See the AI Opportunity Assessment

Sources

  1. OAIC, Australian Privacy Principles guidelines, Chapter 1: APP 1, version 2.0 (30 September 2026)
  2. OAIC, New resources on transparency for use of AI and automated decision-making (30 September 2026)
  3. OAIC, Australian Community Attitudes to Privacy Survey 2026: digital technologies
  4. OAIC, Privacy compliance sweep to put privacy policies under the spotlight
  5. Allens, Mandatory automated decision-making disclosure requirements are coming: key takeaways from the OAIC guidance (2 October 2026)
  6. Office of the Privacy Commissioner (New Zealand), IPP 3A
  7. Office of the Privacy Commissioner (New Zealand), Artificial intelligence and the Privacy Act
  8. OAIC, Rights and responsibilities under the Privacy Act
  9. Privacy and Other Legislation Amendment Act 2024 (Cth), Schedule 1, Part 15

Questions people ask us

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

From 10 December 2026, an organisation covered by Australia's Privacy Act that uses a computer program to make, or substantially and directly help make, decisions that could significantly affect people must describe three things in its privacy policy: the kinds of personal information the programs use, the kinds of decisions made solely by a program, and the kinds of decisions where a program does something substantially and directly related to the decision. This is set out in the OAIC's guidelines on Australian Privacy Principle 1, updated on 30 September 2026.

Does the automated decision rule apply if a person makes the final decision?

Yes, it can. The OAIC's guidelines say a computer program can be captured even where its output is reviewed by a person, and that it is captured where the human decision-maker relies on or is significantly influenced by that output. The factors include how much the person relies on the output, whether they can and do override it, and how deeply the program is built into the workflow.

Does the automated decision rule only apply to AI?

No. The OAIC says the term computer program takes its ordinary meaning and includes rule-based processes as well as machine learning and generative AI tools. The law firm Allens reads the guidance as reaching everyday software such as spreadsheets. The duty sits with the organisation that arranged for the program to be used, even when a vendor supplied it.

Start with a free discovery session.

Thirty minutes with Tom on where software already shapes decisions in your organisation, and what a new agent would add to that list. You leave with a clear view of what is worth automating, what it would take, and what it would return. No obligation.

We use the details you give us only to arrange and hold your session. See our privacy policy.