Compliance · Australia
A practical 2026 guide to the Privacy Act 1988, the Australian Privacy Principles, the 2024 reforms and OAIC enforcement — written for organisations rolling AI into day-to-day work.
AI tools now sit inside almost every Australian organisation — drafting emails, summarising client calls, analysing spreadsheets, triaging support tickets. The Privacy Act 1988 (Cth) still governs every piece of personal information that flows into them. Most practices we work with know this in principle; few have a written framework that stands up to an OAIC review.
This guide gives you that framework. It explains how the Australian Privacy Principles apply to generative AI, what the 2024 reforms changed, what the OAIC expects, and the five steps we recommend before you roll AI out across your organisation.
The short answer. Treat every AI prompt as a disclosure of personal information. Confirm your APP 8 cross-border position, use a paid business plan with a data processing agreement, update your APP 1 privacy policy, train staff on what not to paste, and document the lot. Do that and you have a defensible programme.
The APPs in Schedule 1 of the Privacy Act bind APP entities — most organisations with annual turnover over A$3 million, health service providers of any size, credit providers, TFN recipients and Commonwealth agencies. Four APPs do most of the work when you deploy AI.
Your privacy policy must honestly describe how you handle personal information, including through AI systems. If staff paste client emails into Claude or ChatGPT, your policy needs to say so — what tool, what categories of information, what the vendor does with it, and whether data leaves Australia. A boilerplate policy written before generative AI existed will not meet APP 1 in 2026.
Personal information collected for one purpose can only be used or disclosed for that purpose, unless an exception applies (consent, a related secondary purpose the individual would reasonably expect, or a legal requirement). Feeding a customer's complaint into an AI tool to draft a reply is usually within the primary purpose; using the same data to train an internal model is a secondary purpose that likely needs consent.
This is the one that catches organisations out. More on it below.
You must take reasonable steps to protect personal information from misuse, interference, loss and unauthorised access. With AI, that means vendor due diligence, access controls, retention limits, prompt-level data minimisation and an incident response plan that covers AI-specific failure modes — prompt leakage, model output containing another customer's data, credential theft from a connected tool.
Almost every mainstream AI tool processes data outside Australia. Anthropic's Claude is hosted primarily in the United States. OpenAI's ChatGPT does the same. Google's Gemini uses global infrastructure. When you send a prompt containing personal information, you are disclosing that information to an overseas recipient.
APP 8 requires you to take reasonable steps to ensure the overseas recipient does not breach the APPs. In practice that means one of two paths:
If you rely on contractual equivalence and the overseas recipient mishandles the information, APP 8.1 can make your organisation accountable for the breach as if you had done it yourself. Vendor diligence is therefore not a formality.
The Privacy and Other Legislation Amendment Act 2024, passed in December 2024, is the first tranche of the Attorney-General's privacy reform programme. The provisions most relevant to AI users:
Further tranches — including a possible "fair and reasonable" overarching test and tighter small business exemption rules — remain in consultation through 2026.
The Office of the Australian Information Commissioner published two sets of AI guidance in late 2024 that remain the authoritative reading for Australian organisations:
The OAIC's position is not that AI is off limits. It is that the APPs apply in full, that reasonable steps for AI vendors are higher than for lower-risk suppliers, and that organisations should document their analysis.
Part IIIC of the Privacy Act requires notification of eligible data breaches — where personal information is accessed or disclosed without authorisation and this is likely to result in serious harm to affected individuals. AI tools create new breach vectors:
You have 30 days to assess whether a suspected breach is notifiable, and must notify the OAIC and affected individuals as soon as practicable once you conclude it is. Document your assessment either way — the OAIC frequently asks to see it.
Five steps, in order, before you scale AI across your team:
None of this is onerous. It is the same programme a well-run organisation would build for any new category of vendor — applied with the understanding that AI prompts are a disclosure, not a search.
The Privacy Act does not stop Australian organisations using AI. It asks them to do what they should be doing anyway: know where the data goes, contract for protection, tell people what you are doing, and secure the pipeline. Organisations that get this right will deploy AI faster and with less friction than those that treat compliance as an afterthought — because they will have the governance evidence their clients, insurers and regulators increasingly ask for.
If you want help building that framework, our implementation services include a privacy-by-design review as part of every rollout.
FAQ
Does the Privacy Act 1988 apply to AI tools like Claude and ChatGPT?
The Act applies to your organisation's handling of personal information, including when you send it to an AI tool. If you are an APP entity (most organisations with turnover over A$3 million, plus health service providers and others), the Australian Privacy Principles apply to every input and output you process through an AI system.
Is sending personal information to Claude or ChatGPT a cross-border disclosure under APP 8?
In most cases, yes. Anthropic and OpenAI process data on overseas servers, so sending personal information to them is a disclosure to an overseas recipient. Under APP 8 you must take reasonable steps to ensure the recipient complies with the APPs, or rely on an exception such as informed consent.
What did the Privacy and Other Legislation Amendment Act 2024 change?
The 2024 reforms introduced a statutory tort for serious invasions of privacy, a Children's Online Privacy Code, expanded OAIC enforcement powers including new civil penalty tiers, and transparency obligations for automated decisions that significantly affect individuals. Further tranches of reform are expected through 2026.
Do we need to report an AI-related data breach?
Yes, if the breach is likely to result in serious harm and you cannot contain it, the Notifiable Data Breaches scheme requires you to notify the OAIC and affected individuals as soon as practicable. This includes incidents where personal information was inadvertently disclosed through an AI tool or where an AI vendor suffers a breach.
Can we use free consumer AI tools for work?
We recommend against it for anything containing personal information. Free tiers of ChatGPT, Claude and Gemini typically train on user inputs. Paid business plans — Claude for Work, ChatGPT Enterprise, Gemini for Workspace — contractually exclude training and provide the data processing terms you need for APP compliance.
What does the OAIC expect organisations to do before deploying AI?
The OAIC's 2024 guidance expects a documented privacy impact assessment, a clear lawful basis for each category of personal information processed, updates to your APP 1 privacy policy, vendor due diligence on APP 8 cross-border requirements, and governance over how staff use AI day to day.
We help Australian organisations roll Claude out with the governance, training and documentation the Privacy Act expects — not the box-ticking kind, the kind that actually holds up.