Australia has no AI Act, but the Privacy Act already covers every AI tool that touches personal information. From 10 December 2026, automated decisions also have to appear in your privacy policy. Here's AI Lab Australia's plain-English guide to who's covered, which APPs matter, and the checklist we'd work through first.
Most Australian businesses already use AI on personal information, whether they meant to or not. The Privacy Act has applied to that the whole time. On 10 December 2026 a new rule lands that makes some of it public. Here's our plain-English read of the Privacy Act and AI in Australia, and the short list of things we'd fix first.
If you came here looking for "the new AI law", there isn't one. Australia still doesn't have an AI Act. What it has is the Privacy Act 1988, the Australian Privacy Principles (APPs), and a regulator that has been very clear about how the two fit together.
The Office of the Australian Information Commissioner (OAIC) puts it in one line in its guidance on commercially available AI products: "The Privacy Act applies to all uses of AI involving personal information."
That's the whole game. If a person's details go in, or come out, the APPs apply.
We build AI agents for accounting, legal, finance and health admin teams, so we read this stuff for a living. What follows is our view as an agency, not legal advice.
This is where a lot of small businesses switch off too early.
The Act covers organisations with annual turnover above $3 million. Below that, most businesses are exempt. But the exceptions are wide, and the OAIC's small business page lists them. Regardless of turnover, you're covered if you are, among other things:
Turnover means income from all sources, and the test looks back at any financial year since 2002.
Two more things the exemption doesn't protect you from.
The first is the statutory tort for serious invasions of privacy, which started on 10 June 2025. The OAIC notes it "extends to individuals and other entities that may not necessarily be an Australian Privacy Principle entity". It targets intentional or reckless conduct, so a careless prompt won't land you in court. A system built to mine client data in ways nobody would expect might.
The second is your clients. If you work for businesses that are covered, expect their supplier questionnaires to ask how you handle data in AI tools.
Our blunt advice: if you're anywhere near health, finance, legal or government work, act as if you're covered. It's cheaper than finding out the hard way.
You don't need all 13 APPs memorised. For AI work, six do most of the heavy lifting.
APP 1, open and transparent management. You need a clear, current privacy policy and the internal practices to back it up. From 10 December 2026 it also carries the new automated decision rules (more below).
APP 3, collection. This is the one people miss. The OAIC says that when AI is used to generate or infer personal information, that counts as a collection. A model that guesses a customer's age, income bracket or health status from their emails is collecting personal information, and has to meet the same "reasonably necessary" and "lawful and fair" tests as a form would. Inferring sensitive information generally needs consent.
APP 5, notice. If an AI vendor can access the personal information you put through its system, the OAIC treats that as a disclosure that belongs in your collection notice.
APP 6, use and disclosure. You can use personal information for the purpose you collected it for. Pasting a client's file into a public chatbot is a disclosure to the chatbot's owner, and probably not what the client signed up for. The OAIC's best-practice position is direct: don't enter personal information, and especially sensitive information, into publicly available generative AI tools.
APP 8, cross-border disclosure. Most AI services process data overseas. Under APP 8, you generally stay accountable for what an overseas recipient does with information you send it. "The vendor's terms said it was fine" is not a defence you want to test.
APP 10 and APP 11, accuracy and security. AI gets things wrong with total confidence. The OAIC says hallucinated or inferred details about an identifiable person are still personal information, and you have to take reasonable steps to keep personal information accurate. Then APP 11 asks whether connecting a tool to your file server just widened your attack surface.
None of this is new law. It's the old law, applied to new tools.
This is the date to put in the calendar.
The Privacy and Other Legislation Amendment Act 2024 added new transparency rules to APP 1, with a 24-month run-in. They start on 10 December 2026. The OAIC published its fact sheet, flowchart and updated APP 1 guidelines on 30 September 2026.
In short, you're caught by APP 1.7 if all three are true:
If you are, your privacy policy has to say what kinds of personal information those programs use, and what kinds of decisions are made by them, either outright or with them substantially informing a human.
Three details in the APP 1 guidelines are worth reading twice.
"Computer program" is broad. It includes rule-based automation, machine learning and generative AI, chatbots included. Your Zapier or n8n flow that auto-declines refunds over a threshold counts as much as anything with "AI" in the name.
A human in the loop doesn't get you out. The OAIC's example is a scholarship panel that receives an AI ranking and can ignore it. That ranking is still "substantially and directly related" to the decision. If your people rely on the output, it's in scope.
Using a vendor's tool still counts as you arranging it. The OAIC's own examples include buying an AI system to screen and rank job applicants, letting staff use an AI chat tool to draft performance reviews that decide promotions, and contracting software to approve or decline refunds automatically.
Where we'd expect it to bite for the firms we work with: tenant or applicant screening, credit and lending pre-assessment, insurance triage, job applicant ranking, complaint escalation, and any agent that can refuse, suspend or prioritise someone. Drafting emails or tagging invoices in Xero usually won't hit the "significantly affect" bar. If you're unsure, the OAIC's advice is to take a cautious approach and disclose.
The penalty side is real too. The OAIC can now issue infringement notices for a privacy policy that's missing required content (APP 1.4), at up to 60 penalty units per contravention for a company, which the OAIC currently lists as $21,840. That's for the paperwork failure alone, before anyone asks how the system actually behaved.
The OAIC received 1,205 data breach notifications in calendar 2025, the highest since mandatory reporting began in 2018. Health service providers were the most affected sector, with 225. Legal, accounting and management services were in the top sectors with 81. The January to June 2026 figures, released on 26 October, show 667 more, with health again on top.
More personal information flowing through more tools, in the industries that already leak the most. That's why security is our default rather than an add-on.
This is the order we'd work through it in. The first five are quick wins.
If you want a structured way to score where you are, our AI readiness framework for Australian businesses covers the governance layer alongside the technical one.
"We're under $3 million, so it doesn't apply." Read the exceptions list again. Health service providers are covered at any turnover, and that includes gyms and childcare centres.
"The vendor is compliant, so we are." Vendor compliance covers the vendor. Under APP 1.7 the OAIC expects the obligation to sit with the business using the program to make decisions. Your vendor should give you the information, but your policy has to say it.
"A human signs off, so it's not automated." See the scholarship example. Rubber-stamping an AI recommendation doesn't take it out of scope.
"We'll deal with it when the AI law arrives." The Attorney-General's Department ran a consultation on an exposure draft, the Privacy Amendment (Personal Data Protection) Bill 2026, which closed on 18 September 2026. The government describes it as still subject to further consideration. Nobody knows the final shape. What we do know is what applies today and on 10 December.
"Self-hosting fixes everything." The OAIC says on-premises deployment is "likely to be more privacy-preserving", and it eases APP 8. It doesn't fix APP 3, APP 10 or APP 1.7. A badly designed system on a Sydney server is still badly designed.
Is there a specific AI law in Australia? Not as of October 2026. AI use is regulated through existing laws, chiefly the Privacy Act 1988 and the APPs, plus consumer, anti-discrimination and sector rules. The National AI Centre's Guidance for AI Adoption is voluntary.
Can my staff use ChatGPT with client information? The OAIC recommends against entering personal information, particularly sensitive information, into publicly available generative AI tools. A business plan with training disabled, clear terms and an internal policy is a different risk profile, but you still need to check APP 6 and APP 8.
What changes on 10 December 2026? If a computer program makes or substantially informs decisions that could significantly affect people, and uses their personal information, your privacy policy must describe the kinds of information used and the kinds of decisions involved (APP 1.7 to 1.9).
Does a website chatbot need a privacy notice? If it collects personal information, APP 3 and APP 5 apply like any other collection point. The OAIC also says public-facing AI tools should be clearly identified as AI, so people know they're not talking to a human.
Do I need a Privacy Impact Assessment for AI? It's not mandatory for most private businesses, but the OAIC recommends one as part of a privacy-by-design approach, and for high-risk uses it's the most defensible record you'll have.
If you want help, we do two things here. Our AI consulting engagements start with exactly this audit: tools, data flows, APP 1.7 exposure, and a policy your team will actually read. And for firms that can't send client data offshore at all, we build private, Australia-hosted AI that answers from your own documents.
If you've got a question about your setup, email us at info@ailabaustralia.com. We'll tell you straight what we'd fix first, even if the answer is "nothing yet".
This article is general information, not legal advice. It reflects the law and OAIC guidance as of 28 October 2026.
Get expert guidance on implementing ai security solutions for your Australian business
