AI in family law practice: where it saves hours and where risk sits

By Michael Nadalin, Founder, Market Lead · 2 October 2026 · 8 min read

Picture a property settlement file on a Thursday afternoon. Four years of bank statements for three accounts, a self-managed super fund, a family trust with its own financial statements, two sets of tax returns and a client who has sent 140 emails, about a third of which contain something relevant. The trial notice is out and the case outline is due. Most of the work ahead of you is not legal judgment. It is reading, sorting and summarising.

That is the part of family law practice where AI tools are useful right now. It is also where people get sloppy, because the same tool that summarises a bank statement in seconds will just as confidently summarise one it has misread, cite a provision that was repealed in 2024, or sit on a server you know nothing about holding a child's medical history.

This piece is about the split. Where AI saves real hours on a family law file, where it does not, and where the risk actually sits, so you can decide what goes into the tool and what stays on your desk.

Where the hours actually go on a family law file

Family law files are document-heavy in a way that suits AI, and people-heavy in a way that does not. The ongoing duty of full and frank disclosure under the Family Law Rules 2021 means you are receiving, checking and chasing financial documents for the life of the matter. Affidavits run long. Clients send narrative, not evidence. A lot of the work is turning a pile of material into something a judicial registrar or judge can read.

The useful question is not whether AI can do family law. It is which tasks are mainly about volume and structure, where a first draft by a machine and a check by you beats doing it from scratch, and which tasks depend on judgment about people, risk and credibility, where a machine draft gives you false comfort.

Disclosure and financial documents: the real time saver

This is where most of the honest time savings sit. A capable tool can take a folder of bank statements and produce a list of transactions over a set amount, transfers between accounts, regular payments to an unknown party and gaps in the date range. That gives you a working list for a disclosure letter or a subpoena in a fraction of the time it takes to read every page.

The failure mode is specific: optical character recognition on scanned statements. A misread figure, a dropped minus sign or a page read twice will flow straight into your schedule of assets and liabilities if nobody checks it. Treat AI output on financial material as a lead list, not a finding. Anything that ends up in a balance sheet, a financial statement or an affidavit gets checked against the original document by a person.

There is a second trap. AI is good at finding what is in the documents and useless at noticing what is missing. The account the other party never disclosed, the trust distribution that does not appear, the business that stopped trading the month before separation. Spotting those gaps is still your job, and it is often where the money in a property matter actually is.

Parenting matters: the risk is different

Parenting files contain the most sensitive material most solicitors will ever hold: notifications from child protection agencies, medical and counselling records, police reports, family reports and the child's own words. The time savings are real here too, mainly in chronologies and in organising subpoenaed material. The risk is not just a wrong answer. It is the wrong person seeing the material, or a summary that flattens a safety concern into a neutral sentence.

Summaries are where this bites. An AI tool asked to summarise a set of messages between separated parents will tend to produce something balanced and calm. Sometimes the right reading of those messages is that one party is engaged in a pattern of coercive control, and a balanced summary hides exactly the thing the court needs to see. Under the best interests framework in section 60CC, safety sits at the front of the analysis. Any tool output on family violence material needs to be read with that in mind.

The law changed, and many AI tools have not caught up

Family law has had two significant rounds of amendment in quick succession. The parenting changes that commenced in May 2024 removed the presumption of equal shared parental responsibility and replaced the old primary and additional considerations in section 60CC with a simpler list of best interests factors. The property changes that commenced in June 2025 put the steps of the property framework into the Act and added express reference to the economic effects of family violence.

General purpose AI tools learn from a large body of text, and much of what has been written about Australian family law predates these changes. Ask one about parenting orders and it may confidently explain a presumption that no longer exists, or describe section 60CC in terms that were repealed. That is not a rare edge case. It is the expected behaviour of a tool that has more old commentary than new.

The fix is simple and dull: verify any statement of law against the current Act on the Federal Register of Legislation or a current authoritative source, and confirm any case before you cite it. Australian courts have already dealt with practitioners who filed material containing citations that did not exist. In a jurisdiction where the court is deciding where a child lives, there is no tolerance for that.

Confidentiality, privilege and where the data goes

The Family Law Act restricts publishing material that identifies parties, children and witnesses in family law proceedings. Uploading a client file to a consumer AI tool is not publication in the ordinary sense, but it is disclosure to a third party you have not vetted, and in family law that third party may hold information about children, health and family violence. Your duties of confidentiality under the conduct rules apply regardless of how convenient the tool is.

Before any family law material goes into an AI system, you should be able to answer a few plain questions. Where is the data stored and is it in Australia? Is it used to train the model? Who at the provider can access it? How long is it kept, and can you delete it? If the answer to any of those is that you do not know, the file should not go in. A sensible approach is to treat consumer chatbots as off limits for client material and only use tools your practice has assessed and approved.

A workable policy for a family law practice

You do not need a 30 page AI policy. A family law practice needs a short list of approved tools, a clear rule that consumer chatbots never see client material, a requirement that every proposition of law and every figure is checked against source, and a sign-off by a practitioner who has read the underlying documents. Add a line about extra care with parenting material and family violence evidence, and you have covered most of the real risk.

Then use the tools where they pay off: indexing disclosure, building chronologies, producing first-pass transaction lists and drafting the structure of correspondence and outlines. Keep the judgment work, risk assessment, credibility, negotiation strategy and anything a child's safety turns on, with the people who are accountable for it.

For legal research, the test is whether the tool shows its working. Legal Brain is built for Australian practitioners and links answers back to the sources they rely on, so you can check a statement about the current Family Law Act yourself instead of taking it on trust. Whatever you use, that is the standard to hold it to.

Frequently asked questions

Can family lawyers in Australia use AI?

Yes, but the professional duties do not change. Solicitors remain responsible for confidentiality, for the accuracy of anything filed and for competent advice. Some Australian courts have issued practice notes on generative AI, so check the current requirements of the court you are appearing in before relying on AI for anything that will be filed.

Is it safe to upload family law documents to ChatGPT?

Uploading client material to a consumer chatbot is a confidentiality risk, especially in family law where files contain information about children, health and family violence. Only use tools your practice has assessed, where you know where the data is stored, whether it is used for training and how it can be deleted.

What family law tasks can AI help with?

The strongest uses are indexing disclosure, building chronologies from client material, producing first-pass lists of transactions from bank records and drafting the structure of letters and outlines. All of these still need checking against the source documents by a practitioner.

Why does AI get Australian family law wrong?

Many AI tools learned from material written before the 2024 parenting amendments and the 2025 property amendments. As a result they may describe the repealed presumption of equal shared parental responsibility or the old section 60CC considerations. Always check propositions of law against the current Act.