The Australian Privacy Act and AI tools: a guide for small businesses
What the OAIC expects when your business uses ChatGPT-style tools or automated decisions, and the new privacy policy rule that starts on 10 December 2026.
Someone in your office pastes a customer complaint into a chatbot to get a polite reply drafted. Someone else uses an AI tool to shortlist job applicants, and the bookkeeper has a spreadsheet that flags which customers get offered a payment plan. None of this feels like "doing AI", but under Australian privacy law each one can be a use of personal information with rules attached.
The Office of the Australian Information Commissioner (OAIC) has published guidance on exactly this, and a new transparency rule about automated decisions starts on 10 December 2026. Here's what a small business needs to know, written in plain terms. It's general information rather than legal advice, so check anything that matters with a privacy lawyer.
First, does the Privacy Act apply to you?#
Many small businesses assume they're exempt, and many are. According to the OAIC's rights and responsibilities page, organisations with an annual turnover of more than $3 million have responsibilities under the Privacy Act. Below that threshold you can still be covered, for example if you're a private sector health service provider, if you sell or buy personal information, if you're a credit reporting body, or if you're a contracted service provider under an Australian Government contract.
If you're covered, you're an "APP entity" and have to follow the Australian Privacy Principles (APPs). If you're not, the guidance below is still a sensible standard to hold yourself to. Check your status properly before you rely on the exemption, because the exceptions are easy to overlook.
The OAIC's view of everyday AI tools#
In October 2024 the OAIC published guidance on privacy and the use of commercially available AI products, updated in January 2025. It's aimed at "deployers", which it defines broadly as any individual or organisation that supplies or uses an AI system to provide a product or service. Using a tool internally counts.
The headline recommendation is blunt. As a matter of best practice, the OAIC recommends that organisations "do not enter personal information, and particularly sensitive information, into publicly available generative AI tools, due to the significant and complex privacy risks involved."
That's a recommendation, not a ban. But it's the regulator's stated position, and it's the line you'd be measured against if something went wrong. In practice it means a staff policy that says: no customer names, contact details, health information or anything identifying goes into a free or consumer AI chatbot.
The guidance also explains why. Some commercial AI products have terms or settings that let the provider collect what customers type in for further training and development of AI. Before you approve a tool, read the terms of service and the data settings to see whether the provider can access what you put in and what it produces. Check the terms for the version and plan you actually use, not the one you read about.
How the privacy principles map onto AI#
The guidance walks through several APPs. These are the ones that come up most for a small business.
Start with purpose. Under APP 6 you can generally only use or disclose personal information for the purpose you collected it for, unless an exception applies, such as consent or a related secondary purpose the person would reasonably expect. The OAIC notes that feeding existing customer data into an AI system can be hard to justify on that basis, and where it isn't clearly established, it suggests getting consent or offering a meaningful opt-out. If you collected email addresses to send invoices, running them through an AI tool to build marketing profiles is a different purpose.
The next point is the one people miss. Under APP 3, if an AI tool generates or infers information about an identifiable person, the OAIC treats that as a collection of personal information. That includes inferences and wrong information, such as hallucinations about a real person. The generation has to be reasonably necessary for your functions and done by lawful and fair means, and for sensitive information you'll usually need consent.
APP 10 covers accuracy. You must take reasonable steps to keep personal information accurate, and the guidance says plainly that AI systems are known to produce inaccurate or false results. Its practical suggestion is to record when something in your files is an AI output, so later readers treat it as a probabilistic assessment rather than a fact. If a tool summarises a phone call into a customer record, mark the summary as AI-generated.
On transparency (APP 1 and APP 5), your privacy policy should be clearly expressed and up to date, and the guidance recommends it includes clear information about how you use AI. It also says public-facing AI tools, such as a chatbot on your website, should be clearly identified as AI to the people using them.
Security and overseas disclosure matter too. APP 11 requires reasonable steps to protect personal information, and the guidance asks you to weigh that when choosing an AI tool. It also flags cross-border disclosure (APP 8) as a consideration when personal information goes into a chatbot, so find out where the tool processes and stores data.
Choosing a tool: the questions to ask first#
The OAIC's checklist for selecting an AI product is two pages long and worth printing. Its questions include whether the system has been tested for your intended use, whether you understand its limitations, what security measures the developer has in place, whether the product is hosted in the cloud, and whether the terms let the developer access personal information you put in or the system generates. If they do, it asks whether you can opt out or put controls around that use.
The main guidance adds that due diligence "should not amount to a 'set and forget' approach". Vendors change their terms and features, so review tools periodically, and train staff on what can and can't go in. The guidance also recommends a privacy impact assessment as part of a privacy-by-design approach. For a small business, that assessment can be a short written note: what data goes in, where it goes, who can see it, what could go wrong and what you've done about it.
The new rule for automated decisions from 10 December 2026#
The Privacy and Other Legislation Amendment Act 2024 added a transparency obligation to APP 1, and the OAIC announced on 30 September 2026 that it takes effect on 10 December 2026.
It applies when all three of these are true: you've arranged for a computer program to make a decision, or do a step in one; the decision could reasonably be expected to significantly affect someone's rights or interests; and personal information about that person is used in running the program for that decision.
If so, your privacy policy needs to describe the kinds of personal information the programs use, the kinds of decisions made solely by them, and the kinds of decisions where a program does a step substantially and directly related to the decision.
The OAIC's fact sheet on the obligation clears up two common misunderstandings.
"Computer program" doesn't just mean AI. The fact sheet lists pre-programmed rule-based processes and ordinary software alongside machine learning, and one of its worked examples is a spreadsheet that scores people to decide who gets offered support. Your Excel formula can be in scope.
A human in the loop doesn't automatically take you out of it either. The obligation can still apply when the program's output is reviewed by a person, if that output is a key factor in the decision. The fact sheet's examples of decisions that can significantly affect people include recruitment software that sorts candidates, programs that approve or reject loan or credit applications, eligibility assessments for insurance, and AI-generated reports used to rank employee performance or set bonuses.
For a small business, the likely candidates are hiring tools that screen or rank applicants, credit or payment-plan decisions, and personalised pricing on significant purchases. Ordinary uses like drafting emails or summarising meeting notes are much less likely to meet the "significantly affect" test on their own.
What to do this month#
Start with an inventory. List every AI tool and every automated rule your business uses, what personal information each touches, and which decisions about people it makes or feeds into. It usually takes an afternoon, and it's the input to everything else.
Then write a one-page staff rule about what can go into AI tools, check the data terms of the ones you've approved, label AI-generated content in customer records, and update your privacy policy. If any item on your inventory makes or informs decisions that matter to someone's job, money or access to services, that policy update has a deadline: 10 December 2026.
Sources
- Guidance on privacy and the use of commercially available AI products — Office of the Australian Information Commissioner (October 2024, updated January 2025) (checked 2026-10-11)
- Privacy considerations when selecting a commercially available AI product (checklist) — Office of the Australian Information Commissioner (checked 2026-10-11)
- Rights and responsibilities — Office of the Australian Information Commissioner (checked 2026-10-11)
- New resources on transparency for use of AI and automated decision-making — Office of the Australian Information Commissioner (30 September 2026) (checked 2026-10-11)
- APP 1.7-1.9 Transparency Obligation (fact sheet) — Office of the Australian Information Commissioner (September 2026) (checked 2026-10-11)
Dealing with this in your own business?
If you want to use AI without sending sensitive data to outside services, tell us what you're working with and we'll suggest a setup.