Notes

AI for Law Firms in Australia, the File Work Not the Advice

How can a law firm use AI?

The safest and most useful use of AI for law firms is the file work around the legal work: client intake, chasing documents, searching what the firm has done before, and preparing first drafts for a lawyer to review. The advice, anything filed with a court, and anything you sign stay with a practitioner.

In Australia that line is not just caution, it is what the regulators have asked for. Their joint statement on AI is clear that the duty to give accurate legal information sits with the lawyer, not the tool. So the useful question is not whether AI can practise law. It is which hours between a new enquiry and your final review a machine can carry.

Which law firm tasks should I automate first?

Start with work that is high volume, follows rules and is already checked by a person. In most firms that means one of five lanes.

The intake, chasing and admin lanes sit inside the admin disappears, wired into the practice software you already use, with a person approving anything that leaves the office. Pick the lane that loses the most hours with the least judgement in it, measure it, and build that first.

What do Australian legal regulators say about AI?

The Law Society of NSW, the Legal Practice Board of WA and the Victorian Legal Services Board and Commissioner issued a joint statement on the use of AI in Australian legal practice in December 2024. It is not a ban. It explains how the duties you already carry apply when AI is involved.

The points a firm can build around:

That shapes how we build. Data stays in accounts the firm owns, the build logs what it touched and who approved it, and nothing advises or files on its own. How the statement applies to your practice, your costs agreements and your staff is a question for your law society or your own risk adviser, not a software supplier.

Court documents are where AI does the least

Treat court documents as the place AI does the least. Courts have set their own rules, and they are stricter than general practice guidance.

The Supreme Court of NSW, for example, has a practice note on generative AI in force since February 2025. It limits the use of generative AI in preparing affidavits, witness statements and expert reports, and restricts putting confidential or suppressed material into these tools. Other courts and tribunals have issued their own notes. Check the current practice notes of every jurisdiction you appear in before anyone on your team uses AI near a filed document.

For a build, court work stays out of scope by default. The machine keeps the file organised and the dates visible. It does not draft evidence.

Can AI search our precedents and past matters?

Yes, and for many firms this is the most valuable build there is, because it is knowledge you already paid for and currently cannot find.

A build over your own precedents, past advice and file notes can find the three matters where you dealt with this issue before and pull the clause your senior associate drafted years ago. Every answer cites the document it came from, so the lawyer reads the source rather than trusting a summary. That is what agents on your data has to mean in a firm: if it cannot cite it, it does not say it.

What it does not do is tell you what the law is today. Your own files can be out of date, and general models have produced case citations that do not exist. A lawyer checks every authority against a primary source.

What about privilege and client confidentiality?

Build so the question never comes up: client material stays in systems the firm controls, and never goes into a public tool.

Whether disclosing privileged material to a third party service puts confidentiality, and so privilege, at risk is a legal question we are not going to answer for you. The safer design avoids it. Access is limited per matter, the build runs in accounts you own, every action is logged, and you can show a client or a regulator exactly what happened. The Australian Privacy Principles sit underneath that, and AI data security for Australian businesses covers the practical safeguards.

Trust money is the other bright line. It is tightly regulated, so a build can report on it but does not move it.

A sensible first project

One lane, one team, thirty days, with the numbers written down before you start. Document chasing is the usual first choice, because the volume is high, nobody enjoys it, and the result is easy to count.

Take the baseline first: how many matters are waiting on client documents, how many reminders a typical matter needs, and how many staff hours go on chasing each week. Then judge it honestly at thirty days. If you run a broader practice, AI for professional services firms covers the billable hour question that sits underneath all of this.

Frequently asked questions

Will AI replace lawyers?

No. It takes the intake, chasing, searching and first drafting, which frees lawyers for advice, advocacy and client relationships. The regulators are explicit that accuracy and judgement stay with the practitioner, so the lawyer cannot be designed out.

Can I put client documents into ChatGPT?

Not into a public tool, no. The regulators' statement puts client confidentiality first, and the OAIC's guidance on commercially available AI products warns that what you enter may be stored or used. A build that runs in accounts the firm controls is the safer path.

Do I have to tell clients we use AI?

The regulators' statement expects transparency about AI use, and recording and disclosing it where appropriate. How that lands in your costs agreement and client communications is a question for your law society or risk adviser.

See it before you commit

If you would like to see the crew work before you talk to anyone, the test bench lets you put questions to them on a staged business. Ask something awkward and watch whether it cites its source or admits it does not know. That is the behaviour a law firm needs.

Keep reading