Australia: the Privacy Act and AI in law firms
By the ChamberOne team 4 minute read
An Australian firm that pastes a client's file into an AI tool is handling personal information, often sensitive information, and sending it to a third party. Whether the Privacy Act applies to a particular firm depends on its circumstances, and this article is general information, not advice. For firms that are covered, the Australian Privacy Principles give a useful checklist of the questions to ask before any AI tool is switched on.
Key points
- Work out whether the Privacy Act covers your firm instead of assuming.
- Sending client text to an AI vendor is a disclosure; check use, security and overseas handling.
- Get written vendor answers on training, retention and where data is processed.
- Include AI vendors in your breach plan and keep a register of tools.
Start with whether the Act covers your firm
The Privacy Act 1988 applies to many private sector organisations but not to all, and some small businesses fall outside it. Some firms are covered for other reasons, such as handling certain kinds of information or contracting with government. Do not assume either way.
Check with the Office of the Australian Information Commissioner, or take advice, on whether your firm is covered. Even if it is not, clients may expect the same standard, and professional confidentiality duties apply regardless. Check the guidance from your regulator or bar on confidentiality.
Why AI raises privacy questions
Legal files contain names, addresses, health details, financial records and accounts of family breakdown. When an AI tool processes that text, the information is collected, used and often disclosed to the vendor that runs the model. The Australian Privacy Principles, which govern how covered organisations handle personal information, apply to each of those steps.
The OAIC has published guidance on privacy and the use of commercially available AI products, with checklists for choosing and using them. Among other things, it treats it as best practice not to enter personal information into publicly available generative AI tools. It also encourages organisations to check that a product suits its intended use before adoption, and to consider a privacy impact assessment.
The principles that matter most in practice
Without quoting numbers or wording, here are the areas to work through.
- Purpose and notice. Would a client expect their information to be given to an AI service? If your privacy notice and engagement terms do not mention it, consider updating them.
- Use and disclosure. Sending client text to a vendor is a disclosure. Be clear what the vendor may do with it, including whether it is used to train models.
- Overseas disclosure. Many AI providers process data outside Australia. Check where your vendor processes and stores data, and what contractual protections apply. Take advice on what the Act requires of you when information goes overseas.
- Security. Consider how the vendor protects data, who can access it, and how long it is kept.
- Accuracy. AI output about a person can be wrong. Do not rely on it for decisions affecting someone without checking.
- Access and correction. If a person asks what you hold about them, can you find what was sent to the tool and what came back?
Vendor questions to put in writing
Do not rely on a sales presentation. Ask for written answers to questions like these.
- 1
Where is our data processed and stored, including by any sub-processors?
- 2
Is our data used to train or improve models, and can this be turned off?
- 3
How long is data retained, and can we have it deleted?
- 4
Who at the vendor can see our content, and under what controls?
- 5
What happens if there is a security incident, and how will we be told?
- 6
Can AI features be switched off for particular matters or clients?
Notifiable data breaches
Australia has a Notifiable Data Breaches scheme, under which covered organisations may have to notify affected people and the OAIC about certain eligible breaches. The details, including what counts as serious harm and the time limits, should be taken from the OAIC and from advice, not from a blog.
What a firm can do now is prepare. Include AI vendors in your incident plan. Know how a vendor would tell you about a breach, who at your firm would receive the message, and how quickly you could work out which clients' information was involved. If you do not keep a record of what was sent to which tool, that last step is hard.
A practical sequence for a small firm
Map what you intend to use AI for. Classify the information involved: ordinary, sensitive, privileged. Decide which tasks may use an approved tool and which may not use one at all. Assess the vendor with the questions above. Update your privacy notice and engagement terms if needed. Train staff, and ban the use of personal accounts on public tools for client material.
Then keep a short register of tools, what they are used for and when each was last reviewed. Review it at least once a year, or when the vendor changes its terms.
- Map what you intend to use AI for and classify the information
- Assess the vendor with written questions
- Update your privacy notice and engagement terms if needed
- Keep a register of tools and review it at least yearly
- Put personal information into publicly available AI tools
- Rely on a sales presentation
- Let staff use personal accounts on public tools for client material
What this article does not do
It does not say what the Privacy Act requires of your firm in a particular situation. It does not cover state and territory privacy laws, health records laws or the rules of the profession in each state. Those vary, and some will matter more than others depending on your work.
For authority, go to the OAIC and to the guidance of your law society or bar association. Where the stakes are high, take advice.
A hypothetical example
Suppose a three-lawyer Australian family law practice wants AI to summarise long affidavits. The privacy officer asks four questions before approving it. Where is the text processed? Is it used to train models? How long is it kept? Can the feature be limited to matters where the client has agreed?
The answers come back in writing. The vendor processes data overseas and retains it for a limited period, and training use can be switched off. The firm updates its engagement terms to explain this, records the decision in its register and tells staff the tool is approved only for that purpose. None of it is complicated, but all of it has to be done before the first affidavit goes in.
Sources
Controlling where AI is used in ChamberOne
- Choose the model for each AI feature, GPT-5 or Claude, or switch AI features off for the firm.
- With AI features off, nothing is sent to an AI provider.
- Every AI call is recorded in the audit log with the model and token counts.
What it does not do
- With AI features on, the text needed for a request is sent to the AI provider for that feature, outside the country where ChamberOne is hosted.
- It does not decide whether the Privacy Act applies to your firm or make your firm compliant.
- AI output can be wrong or incomplete and needs a lawyer's review.
This article is general information about the law, legal practice and technology, current when it was published. It is not legal advice. Laws and guidance change: check the sources linked above and take advice on your own circumstances from a lawyer qualified where you are.
Read next
Generative AI guidance for lawyers: how six markets compare
How to find what regulators, courts and bar bodies have said in six markets, and the common themes across them.
5 minute readAustralia: court practice notes on generative AI and what they require
Courts in Australia have issued practice notes on generative AI. Read the one for your court, every time.
4 minute readEU AI Act: what it means for firms outside the EU with EU clients
How a firm outside the EU should think about the EU AI Act when it works for European clients.
5 minute read