Australia and generative AI: what the OAIC and the Privacy Act require

Australia has no dedicated law on artificial intelligence. That does not leave its companies in a legal vacuum. The Privacy Act 1988 and its thirteen Australian Privacy Principles (APPs) apply as soon as personal information goes into an AI tool, and in October 2024 the Office of the Australian Information Commissioner (OAIC) set out how it reads those rules when it comes to generative AI.

We looked at this framework, the reforms passed at the end of 2024 and the special status of two very Australian identifiers: the Tax File Number and the Medicare number. Here is what matters.

The foundation: the Privacy Act 1988 and the APPs

The Privacy Act 1988 applies to federal government agencies and to most private organisations with an annual turnover above 3 million Australian dollars, as well as to some small businesses, notably those providing health services. The APPs govern the collection, use, disclosure, quality, security and cross-border transfer of personal information.

Several principles bear directly on the use of AI:

• APP 1 and APP 5: transparency, through an up to date privacy policy and collection notices.

• APP 6: information collected for one purpose cannot be reused for another without consent, unless the individual would reasonably expect it.

• APP 8: accountability for disclosures overseas, which are common with AI services hosted outside Australia.

• APP 11: reasonable steps to protect information from misuse, loss and unauthorised access.

What the OAIC says about commercially available AI products

On 21 October 2024, the OAIC published two guides: one for developers training models, the other on privacy and the use of commercially available AI products. The second guide is aimed at every organisation deploying a chatbot, a writing assistant or an AI feature built into existing software.

Its clearest message concerns public tools: the OAIC advises organisations not to enter personal information, and particularly sensitive information, into publicly available generative AI tools, because of the significant and complex privacy risks involved. The guide also recommends:

1. due diligence on the product, its data flows and its security risks before adoption;

2. minimising the information that is entered;

3. updating privacy policies to explain when and how AI uses personal information;

4. privacy impact assessments and ongoing governance.

The 2024 reforms raise the stakes

The Privacy and Other Legislation Amendment Act 2024 received Royal Assent on 10 December 2024. It is the first stage of a broader overhaul of the Privacy Act. Among its changes:

• a statutory tort for serious invasions of privacy, available since 10 June 2025;

• new transparency obligations for automated decisions, applying from 10 December 2026;

• tiered civil penalties and the power for the OAIC to issue infringement notices directly;

• clarification that the security steps required by APP 11 include technical and organisational measures.

The December 2026 deadline is close. Organisations that use computer programs to make significant decisions about individuals will have to say so in their privacy policy, whether or not those programs rely on AI.

TFNs and Medicare numbers: two identifiers in a class of their own

The Tax File Number has its own regime, the Privacy (Tax File Number) Rule 2015. It prohibits collecting, using or disclosing a TFN outside the purposes allowed by taxation, personal assistance or superannuation law, and requires TFN information to be protected and then destroyed once it is no longer needed. Pasting a staff list with TFNs into a chatbot to tidy up a payroll file clearly falls outside that framework.

The Medicare number is a government related identifier covered by APP 9, and it almost always travels with health information, which the Privacy Act treats as sensitive information. In a GP practice, a clinic or a health insurer, it appears in letters, invoices and files that administrative staff handle every day.

Where the real risk lies

The typical scenario is not an attack. It is a busy employee using a free AI tool, often outside any control, which is what we call shadow AI. An AI data leak usually follows the same path: a copy and paste, an attached file, then data leaving the company. With AI agents and the MCP protocol, these flows multiply, which is why MCP needs its own DLP layer.

ZeusLock works on the device and in the browser, before anything is sent. It spots sensitive data, including identifiers such as TFNs or Medicare numbers, then anonymises it while preserving its format, warns the user or blocks the upload according to your rules. Every incident is logged, which helps demonstrate the reasonable steps APP 11 expects.

Key takeaways

• The Privacy Act 1988 and the APPs apply to any personal information entered into an AI tool.

• The OAIC advises against entering personal information, especially sensitive information, into public generative AI tools.

• Transparency obligations for automated decisions apply from 10 December 2026.

• TFNs and Medicare numbers need stronger protection.

Want to know which sensitive data is actually leaving your devices for AI tools? Book a ZeusLock demo.