Skip to content
Skip to content

Practical automation guide · By Yes AI

Privacy Act checks before AI reads customer data

Published 6 min read

Automations that use AI often read personal information: names in emails, addresses on orders, details in support tickets. The Privacy Act and OAIC guidance set rules for that use. This guide sets out the checks to run before a workflow goes live; it is general administrative guidance, not legal advice, so for your own situation read the OAIC guidance or ask a privacy lawyer.

Quick answer

If your business is covered by the Privacy Act, personal information that an AI workflow reads or produces must be handled under the Australian Privacy Principles. Check that the use fits the purpose you collected the information for (APP 6), what the AI vendor can do with inputs, whether data leaves Australia (APP 8) and how it is secured. From 10 December 2026, privacy policies must also describe certain automated decisions.

Practice manager in a Hobart office reviewing a printed privacy policy beside a laptop with a blurred screen

Does the Privacy Act apply to my business?

The OAIC states that most small businesses are not covered by the Privacy Act, but some are. A small business here means annual turnover of $3 million or less, counting all income from all sources. Several kinds of small business are covered regardless of turnover, and the OAIC recommends that small businesses consider opting in, citing increased customer confidence and trust. Even where the Act does not apply, larger customers may require similar handling in their contracts, so check those too.

  • Health service providers, which the OAIC says include allied health, complementary therapists, child care centres and private schools.
  • Businesses that trade in personal information, such as selling a customer list.
  • Contractors providing services under a Commonwealth contract.
  • Credit reporting bodies.
  • Businesses that have chosen to opt in.

Can we put personal information into an AI tool?

The OAIC states that privacy obligations apply to any personal information input into an AI system and to output that contains personal information. Under APP 6, you may use or disclose personal information only for the purpose you collected it for, unless the person consents or would reasonably expect the secondary use and it is related to the original purpose (directly related, for sensitive information). A secondary use is more likely to be expected if your collection notice and privacy policy described it. As a matter of best practice, the OAIC recommends not entering personal information, particularly sensitive information, into publicly available generative AI tools.

What should we check in the AI vendor's terms?

The OAIC advises reviewing the terms and settings that apply to your use of the product, and in particular whether they give the developer access to data you input or generate. If they do, you need to decide whether using the product complies with APP 6, and what measures to put around it, such as turning off features that disclose data, preventing staff from entering personal information and training them. Business plans from AI vendors often have different data terms from free tools, so read the terms for the plan you will actually use.

  • Whether inputs or outputs are used to train the vendor's models.
  • How long prompts, files and outputs are retained.
  • Where data is processed and stored.
  • Which subcontractors can access the data.
  • How the vendor notifies you of a security incident.

What if the data is processed overseas?

Many AI models run in data centres outside Australia. The OAIC lists cross-border disclosure under APP 8 among the obligations to consider when personal information goes into an AI product, alongside collection under APP 3 and accuracy under APP 10. Find out where each service in the workflow processes and stores data, including the automation platform, the AI model provider and any document extraction service. Some vendors offer processing in Australian regions. Record the locations, reflect them in your privacy policy where required and revisit them when a vendor changes its terms.

How do we limit what the AI sees?

The OAIC expects organisations to consider what information is necessary and to minimise the personal information put into an AI product. Design each AI step to receive only the fields it needs to do its job. Information that has been appropriately de-identified is no longer personal information, according to the OAIC, although whether a person is still reasonably identifiable depends on context. APP 11.3 notes that reasonable steps to secure personal information include technical and organisational measures, so access controls on the workflow itself matter as much as the AI vendor's.

  • Send the order lines, not the whole customer record.
  • Replace names and account numbers with internal reference numbers before the AI step.
  • Exclude health, financial hardship and similar sensitive details unless the task needs them.
  • Limit who can view the workflow's logs, which often contain the same data.

What changes on 10 December 2026?

The Privacy and Other Legislation Amendment Act 2024 added a transparency obligation to APP 1 that starts on 10 December 2026. It applies where you have arranged for a computer program to make a decision, or do something substantially and directly related to making one, that could reasonably be expected to significantly affect an individual's rights or interests, using their personal information. Your privacy policy must then describe the kinds of personal information used and the kinds of decisions involved. Most back-office automations, such as coding supplier bills, do not decide anything about individuals. A workflow that declines credit applications or shortlists job applicants may, so check it against the OAIC fact sheet.

Privacy questions by workflow type (general guidance)

WorkflowPersonal information involvedMain questionsPractical control
Reading supplier invoicesSole trader names and bank detailsAPP 6 purpose; securing the dataLimit access to accounts staff and the workflow account
Drafting replies to customer emailsNames, contact details, order historyAPP 6; APP 8 if the model runs overseasSend only the thread needed; a person reviews before sending
Summarising support ticketsComplaints, sometimes health or financial detailsSensitive information under APP 6Redact before the AI step; exclude sensitive categories
Screening job applicationsWork history and contact detailsAPP 1.7 if a program shapes who progressesA person decides; describe the process in the privacy policy
Approving trade credit accountsFinancial and identity detailsAPP 1.7 automated decisions; APP 10 accuracyA person reviews every decline and records the reason

Practical checklist

  • Whether the business is covered by the Privacy Act has been checked against the OAIC small business tests.
  • Each field of personal information sent to an AI step is needed for the task.
  • The AI vendor's terms on training, retention and data location are recorded for the plan in use.
  • The privacy policy and collection notices describe the AI use where required.
  • Sensitive information is excluded or redacted before any AI step.
  • Any decision about an individual made by the workflow has been assessed against APP 1.7.
  • The data breach response plan names every vendor in the workflow.

Common questions

Our turnover is under $3 million. Does any of this apply?

Possibly. The turnover test exempts many small businesses, but health service providers, businesses that trade in personal information, Commonwealth contractors and some others are covered whatever their turnover. Customers also expect careful handling, and larger clients may require it by contract. The OAIC website has a set of questions that walk through the tests.

Is pasting a customer email into ChatGPT a disclosure?

It can be. The OAIC explains that putting personal information into an AI system is a use if the data stays within your organisation's control, and a disclosure if it becomes accessible to others outside it. For publicly available chatbots, the OAIC recommends as best practice not entering personal information at all. Business plans with contractual data terms need separate assessment.

Do we need a privacy impact assessment?

The OAIC recommends a privacy by design approach for AI products, including a privacy impact assessment. For a small workflow, that can be a short written record of what personal information is used, why, where it goes, who can see it and what could go wrong. Larger or higher-risk uses deserve a fuller assessment.

What if the automation causes a data breach?

Under the Notifiable Data Breaches scheme, organisations covered by the Privacy Act must notify affected individuals and the OAIC when a data breach is likely to result in serious harm. The notice to individuals must include recommended steps they should take. Include your automation and AI vendors in your response plan so you know how they will tell you.

Is information the AI generates about a person also covered?

Yes. The OAIC states that inferred, incorrect or artificially generated information about an identified or reasonably identifiable individual is personal information. Generating it counts as a collection under APP 3, so a workflow that infers details about customers must meet the same rules as one that collects them directly.

Sources and further reading

Continue planning

Turn the guide into your workflow

Bring one process, the systems it touches and the exceptions your staff handle. We can discuss approval boundaries, recovery steps and the evidence needed before changing live operations.