Skip to content
Skip to content

Practical LLM operations guide · By Yes AI

LLMs and personal information: Privacy Act duties

Published 6 min read

If your organisation is covered by the Privacy Act, the Australian Privacy Principles apply to personal information that goes into an LLM and to what comes out. A new transparency rule for automated decisions starts on 10 December 2026. This guide maps the principles to the parts of an LLM project where they apply, as administrative guidance only.

Quick answer

The Privacy Act applies to personal information you put into an AI system and to personal information it generates. In practice: check the use matches the purpose of collection (APP 6), explain AI use in your privacy policy (APP 1), treat inferred information as a new collection (APP 3), take reasonable steps on accuracy (APP 10) and security (APP 11), and assess overseas disclosure (APP 8). From 10 December 2026, privacy policies must also describe certain automated decisions.

Privacy officer and IT lead reviewing a printed data flow diagram at a Canberra office meeting table

Does the Privacy Act cover your organisation?

The OAIC says most small businesses are not covered by the Privacy Act, and defines a small business as one with an annual turnover of $3 million or less, counting income from all sources. Some businesses are covered regardless of turnover, including health service providers, businesses that trade in personal information and contractors providing services under a Commonwealth contract. If you are not covered, your client contracts may still require Privacy Act-level handling, so check those as well. This guide is administrative guidance for planning an LLM project. Confirm your own position using the OAIC's resources or a privacy adviser before relying on it.

Which principles apply when staff use an LLM?

The OAIC's guidance on commercially available AI products sets out how the existing principles apply. Privacy obligations apply to personal information input into an AI system and to output data generated by AI. Each principle below changes a specific design choice, from what goes into the index to how long conversation logs are kept. The OAIC also recommends, as best practice, that organisations do not enter personal information, and particularly sensitive information, into publicly available generative AI tools.

  • APP 1: update your privacy policy and notifications with clear information about your use of AI.
  • APP 3: generating or inferring personal information with AI is a collection and must comply with APP 3.
  • APP 6: use or disclose personal information only for the primary purpose of collection unless an exception applies; the OAIC notes a secondary use may be acceptable if it was expressly outlined in a notice at collection and in your privacy policy.
  • APP 10: take reasonable steps to ensure the accuracy of personal information you collect, generate, use and disclose with an AI product.
  • APP 11: protect personal information from misuse and unauthorised access, and destroy or de-identify it once it is no longer needed.

Can you index documents that contain personal information?

A knowledge assistant over HR files, client correspondence or case notes is a new use of that information. Ask whether answering staff questions from those records fits the purpose the information was collected for. A narrower design often avoids the problem: index the procedure documents and leave the individual files out, or restrict a collection to the team that already handles those records. Apply your retention rules too. If a document should have been destroyed under your retention schedule, it should not survive inside a search index or vector store. A privacy impact assessment is a structured way to work through these questions while the design can still change.

When does APP 8 apply to an offshore AI service?

APP 8 covers disclosure of personal information to an overseas recipient. Where it applies, you must take reasonable steps to ensure the recipient does not breach the APPs, and under section 16C you can be accountable for the recipient's acts. The OAIC's guidelines note that giving information to an overseas cloud provider only for storage and access, under a binding contract that keeps it in your effective control, may be a use instead of a disclosure. Whether an offshore LLM service fits that description depends on its terms, including abuse-monitoring retention and human review. Get advice on your specific arrangement, or keep processing in Australia.

What changes on 10 December 2026?

The Privacy and Other Legislation Amendment Act 2024 added a transparency obligation to APP 1, starting 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, and personal information is used in the program's operation. Your privacy policy must then describe the kinds of personal information used, the kinds of decisions made solely by the program, and the kinds of decisions where the program does something substantially and directly related to the decision. An LLM that screens job applicants or triages claims could fall within it.

The OAIC published new resources on this obligation on 30 September 2026. Read them before deciding whether an LLM workflow is in scope, and record the reasoning either way.

How do you respond if something goes wrong?

An LLM can expose personal information in new ways. A retrieval permission error might show one employee another person's HR record, or a prompt log containing customer details might be shared with a contractor who should not see it. Add these scenarios to your data breach response plan so the assessment starts quickly. The plan should name who can disable a source or the whole assistant, how you will work out which users saw which answers, and how long the relevant logs are kept. Without retrieval and access logs, you may not be able to tell who was affected.

Where each APP shows up in an LLM project

PrincipleLLM design questionTypical control
APP 1 (transparency)Does the privacy policy explain AI use and, from 10 December 2026, relevant automated decisions?Policy update before launch
APP 3 (collection)Will the system generate or infer personal information?Treat those outputs as collected information
APP 6 (use and disclosure)Does answering questions from these records fit the original purpose?Exclude records or narrow the audience
APP 8 (cross-border)Is personal information processed or retained offshore?Onshore processing or a documented assessment
APP 10 (accuracy)Could an answer state wrong facts about a person?Citations and human review for decisions
APP 11 (security and destruction)Do indexes and logs keep data past its retention date?Access controls and synced deletion

Practical checklist

  • You have confirmed whether the Privacy Act covers your organisation, including the turnover test and the listed exceptions.
  • Each source collection is checked for personal information and its original collection purpose.
  • The privacy policy and any collection notices describe the AI use before launch.
  • Offshore processing and retention are identified and assessed under APP 8, or avoided.
  • Workflows that make or substantially inform significant decisions about people are checked against the 10 December 2026 transparency rule.
  • Indexes, vector stores and logs follow the same retention and destruction schedule as the source records.
  • The data breach response plan includes LLM-specific scenarios.

Common questions

Can we put customer records into a private LLM?

Possibly, if the use fits the purpose those records were collected for or an exception applies, and if your privacy policy and notices cover it. A private deployment helps with security and cross-border questions, yet it does not settle the purpose question under APP 6. Start with the narrowest set of records that meets the business need.

Is an LLM summary of a person's file personal information?

Treat it as personal information when it is about a person. The OAIC says privacy obligations apply to output data generated by AI, and that generating or inferring personal information is a collection under APP 3. Store and secure summaries the same way as the source file, and include them in your retention schedule.

Does the automated decision rule apply to every chatbot?

It applies only where a computer program makes a decision, or does something substantially and directly related to making one, that could reasonably be expected to significantly affect an individual's rights or interests, and personal information is used in its operation. An assistant answering staff questions about leave policy is unlikely to meet that test; one that ranks applicants may. Check the OAIC's resources for your case.

Do these rules apply to a small business?

The OAIC says most small businesses with annual turnover of $3 million or less are not covered, but some are covered regardless of turnover, including health service providers and businesses that trade in personal information. Client contracts can also require privacy controls. The OAIC's small business page includes a checklist to confirm your position.

Where can we get a definitive answer?

From the OAIC's guidance and, for your specific facts, a privacy lawyer or adviser. This page is administrative guidance written for planning an LLM project, and it does not cover every exception in the Act. The OAIC's guidance on commercially available AI products is the most direct starting point for organisations using hosted AI services.

Sources and further reading

Continue planning

Plan the checks before your next LLM release

Describe the task, source documents and people who will review the answers. We can discuss evaluation, access boundaries and a staged implementation. Keep confidential documents and access credentials out of this enquiry.