Best AI for Family Law in Australia: Privacy, Court Rules and Cost

Use a closed tool, ideally the AI inside your practice-management system, for anything involving client or court material, and keep free tiers and public assistants for tasks with no client information. The Family Court's 2026 practice direction keeps confidential or sensitive information out of public AI tools, so the shortlist is a choice of closed tools and what they cost.

The practice direction is not the only rule. The Privacy Act applies to the sensitive information a family practice holds every day, and family law adds a child-safety dimension that other practice areas do not have. None of it is legal advice: the court, regulator and vendor material below was read in November 2026, and where I rely on commentary rather than the instrument itself, I say so. Open the current text of anything that affects a decision.

Follow me on Instagram@sagnikteaches

What the Family Court's practice direction asks of a practice

The Federal Circuit and Family Court of Australia issued its practice direction on the use of artificial intelligence on 29 May 2026. It covers lawyers, private experts, litigants and other court users in both divisions of the Court. Its main requirements, read from the direction itself, translate directly into office rules.

Connect on LinkedInSagnik Bhattacharya

• Principles. Integrity, accountability, accuracy, confidentiality and data security, safety, and education. Court users must not use AI where there is a risk it will compromise the safety of any court user or child, such as by exposing a protected address or medical records, and the Courts will not accept a lack of understanding as an excuse for a mistake.
• Wide definition. Obligations apply whether or not a product is described as AI, GenAI, smart or assistive. The direction notes that AI is now built into drafting tools, search, transcription and document management, so a practice cannot say it does not use AI because it never opened a chatbot.
• Public against closed tools. Do not put confidential or sensitive information into public AI tools. Before uploading such material, the user needs to be satisfied, from the tool's terms of service, that it protects the information and restricts its later use. Turn off chat history and web search in a chatbot where that option exists.
• Protected material. A strict prohibition applies to material produced under discovery or disclosure orders, suppression or non-publication orders and subpoenas, unless the information stays in a closed environment under enforceable confidentiality terms, is not used to train any model and is used only for that proceeding.
• Accuracy. Refer only to existing, accurate legislation, case law and sources, and check everything.
• Explaining use. If asked, be able to say whether AI was used, which tool, how the output was checked and supervised, and how the principles were met. No disclosure statement is needed unless the Court asks.
• Experts. Lawyers who engage a private expert give them a copy of the direction and make sure the expert certifies they have read it.
• Consequences. The Court can disregard material, make costs orders or refer the matter to a regulatory body. Improper use that may affect the integrity of proceedings is to be raised with the presiding judicial officer.

The direction is reviewed over time, so the approach will change. A firm should diarise a check of the Court's page every quarter. For the wider Australian picture across courts, the tutorial on AI for Australian lawyers maps the other instruments.

Subscribe on YouTube@sagnikteaches

What reported AI mistakes in the Court teach a practice

The direction was developed by the Courts' AI Committee after consultation, but the Court had already met AI errors in reported cases, and the pattern is useful training material. The details below come from published commentary rather than the decisions themselves, so read the decision before you quote it.

In one reported case, a solicitor handed up a list of authorities produced with AI-assisted research software; some of the cases did not exist, he had not checked them, and the matter was referred to the professional regulator. Commentary on later appeals describes a self-represented party relying on invented case law, and a paralegal using AI to prepare a summary of argument and a list of authorities without the supervising solicitor knowing, with the criticism centred on the missing supervision as much as on the errors.

Two lessons follow. The error is rarely the whole problem; the failure to know who used what is, and the direction now makes court users accountable for the tools they choose and ready to explain how they used them. And confidentiality is a court risk as well as a client one: the direction warns that information entered into a GenAI tool could become public and breach a suppression order or a statutory ban on publication, and it tells court users to respect legal privilege when using these tools. A family practice should treat both as live risks.

Privacy Act duties and the free-tier rule

A family practice holds some of the most sensitive personal information there is: health, finances, family violence history and information about children. The Office of the Australian Information Commissioner's guidance on commercially available AI products says that, as best practice, personal information, and especially sensitive information, should not be entered into publicly available generative AI tools. It also makes these points that matter here.

• Entering personal information into an AI system may be a use or a disclosure, so it needs to fit the purpose you collected it for or a purpose the client would reasonably expect (APP 6).
• You should check where the provider's servers are, since information may be disclosed outside Australia (APP 8).
• You should take reasonable steps to keep the information secure and to check accuracy (APPs 11 and 10).
• Where AI collects or generates sensitive information, consent is needed, and a privacy notice alone is not consent.

The direction adds a family-law point of its own. Under Part XIVB of the Family Law Act it is an indictable offence to communicate to the public an account of proceedings that identifies a party, a witness or a child, and the direction warns that court documents entered into a GenAI tool could become public in exactly that way. Whether a given tool creates the risk turns on how it handles the data, which is one more reason the safe default is closed tools only.

That is the reason for the free-tier rule. Free and personal plans are public tools, their data settings differ and keep changing, and a practice cannot show a client or the Court that information entered there stayed private. Business plans from the main assistant vendors do not train on your content by default, which is why they appear in the shortlist, but they are still third parties, so the closed-environment questions below still apply. For a family-law-specific view of what to pay for, see best legal AI for family law: what to pay for and what to skip.

The shortlist, with prices

These are the tools a small family practice is most likely to compare. Prices are list prices from the vendors' own pages. Australian-dollar figures come from Australian pages, and USD figures are marked as such.

I have left out tools whose Australian pricing or data terms I could not confirm. If a vendor you are considering is not in the table, ask it the five questions below before you trial it.

Five vendor questions drawn from the direction

1. Does the contract bar any use of our inputs and outputs to train a model, the vendor's or anyone else's? The direction's rule on protected material depends on it.
2. Does the data stay in a closed environment under confidentiality terms the vendor is bound by, and who are the sub-processors and where do they store it?
3. Can we switch off chat history and web search across the firm? The direction asks for both where the option exists.
4. Can each matter's data be kept to that matter and deleted on closure? Protected material may be used only for its own proceeding.
5. Does any recording or transcription feature run during hearings? The rules of court bar recording and transcription devices in proceedings, AI tools included, so it must be off.

Ask too how quickly the vendor will tell you about a breach. Get the answers in a document you can file with the privacy impact assessment, and repeat the questions when the vendor changes its terms. What a small practice should check before buying family law AI software gives a longer buying list.

Where AI helps in a family practice, and where it stays manual

Chronologies. A chronology of separation, property events and parenting arrangements is slow to build and easy to get wrong. Inside a closed tool the AI can draft it from the matter's documents. Each entry must link to a source, and a lawyer checks every date. Building chronologies with events linked to evidence goes through the method.

Financial disclosure. Summarising bank statements, superannuation statements and valuations is a good fit for a closed tool and a poor fit for a public one. The summary is a draft: figures are checked against the documents before any schedule is relied on. Summarising financial disclosure safely covers it.

Client emails. A calm, consistent tone matters in a stressful matter. Give the assistant the stage and next steps, not the facts, and edit the output. Family law client communication with AI shows examples.

Intake. A questionnaire that sorts urgency can save hours, but a person must read every response for safety signals, and the tool never decides who is in danger.

What stays manual. Risk assessment around family violence or child safety, advice on parenting arrangements, the content of affidavits, and any judgement about credibility. The practice direction's safety principle is the reason, and so is common sense.

Thirty matters, four lawyers: where the saved hours come from

Illustration: a four-lawyer family practice with 30 active matters trials AI for a quarter on four recurring tasks. The counts and minutes are assumptions to replace with your own trial figures. What matters is the last column, because it decides which tool each task may use.

The sum: chronologies save 17.5 hours a month, disclosure summaries 12.5, emails 7 and intake 2.5, so 39.5 hours in all. At an assumed AUD 110 an hour that is about AUD 4,345 a month, but AUD 3,575 of it (32.5 hours) sits in tasks that only a closed tool may touch. An assistant on its own captures the email time, about AUD 770 a month, and nothing else without breaking the direction's rules or the free-tier rule.

Now the costs. Four Copilot Business seats at the offer price of AUD 26.91 come to AUD 107.64 a month, excluding GST (AUD 125.60 at AUD 31.40 once the offer ends). Four Smokeball Boost seats from AUD 129 come to AUD 516 a month before the Archie AI add-on, but that price buys the whole practice-management system; if the practice already runs it, the real extra cost is the add-on, which Smokeball does not price publicly, so get a quote before the trial. Many practices end up with both, under a written rule on which task goes where.

A prompt, a realistic output and the fix

This shows the sort of task suited to an assistant on anonymised material, with the problems you should expect.

PROMPT (anonymised notes, no names, no dates that identify):
"From these notes about a ten-year relationship and one child, build a table
with columns: approximate date, event, source document. Mark any date you
inferred. Do not add events that are not in the notes."

A typical output (illustrative) is a tidy table of twelve rows. Two problems recur. A row is added saying "separation, which led to the sale of the family home" when the notes mention only the separation: the tool filled in a causal link. And "mid-2019" appears as a date although the notes say "around the time of the move", so the tool picked a year. The fix is to delete the added link, replace each inferred date with the source's own wording, and ask for a "source line" column that the lawyer ticks off against the document. Nothing from the table is used until every row has a tick.

The prompt to refuse is just as instructive. "Turn the client's account of the incident into an affidavit paragraph" asks the tool to generate evidence. Practice notes in other courts prohibit this, and the Family Court's direction stresses that evidence must be the person's own. A lawyer takes instructions and drafts in the client's words.

Telling clients, and briefing experts

Clients in a family matter are anxious about who sees their information, so say it plainly in the retainer, not in a policy they will never open. Here is illustrative wording for discussion with your own adviser.

How we use technology. We use AI features inside our practice-management
system to help draft documents, summarise material and prepare timelines.
These tools run inside our firm's account, do not use your information to
train their models, and are limited to the people working on your matter.
A lawyer checks everything before it is used. We do not use public AI
tools with your information. You can ask us at any time what we have used
on your file.

Experts need the same care, because the practice direction applies to them. When you engage a private valuer or a family consultant, send the direction with the engagement email and ask for the certification to come back with the report.

Subject: Engagement - [matter reference]

Hello [first name],

Please read the Court's practice direction on the use of AI, attached, before
you start. When you send your report, please confirm in writing that you have
read it and tell us whether any AI tool was used in preparing the report and,
if so, which tool and how you checked its output.

[your name]

When it has already gone wrong: the first 24 hours

Suppose a junior pastes a disclosure schedule into a free chatbot to get a quick summary, and mentions it a week later. The direction makes court users accountable for the tools they choose and how they used them, so the response matters as much as the mistake. Within a day: stop further use of that tool on any matter, write down what was entered, when and by whom, and record whether it included material covered by a disclosure order, a suppression order or a subpoena. Tell the supervising lawyer and the practice's privacy officer, then work out with them whether the incident needs to be raised with the client, the Court or the regulator. Take advice before deciding, because the answer depends on the facts and the orders in the matter.

Afterwards, close the gap rather than just reprimanding anyone. Check whether AI features are switched on in software the practice already owns, tighten who can create accounts, and add the incident to training as an anonymised case. A team that sees mistakes handled calmly will report the next one sooner.

A one-page staff rule

AI USE IN [PRACTICE NAME]
1. Approved tools only: [list, with plan]. No client information in any other
   AI tool, including free or personal accounts, and including AI built into
   software that has not been approved.
2. Never enter material produced under disclosure, subpoena, suppression or
   non-publication orders unless the supervising lawyer has approved the tool
   in writing.
3. Evidence is the client's own words. AI does not draft or rework affidavits.
4. Every authority and figure AI produces is checked by a person against the
   source. Checking by asking the AI does not count.
5. Record each use on the file: tool, task, checker, date.
6. Safety, risk and advice on children stay with a lawyer.
7. Tell the supervising lawyer at once if something was entered by mistake.

Rule 7 matters most. A practice that makes it safe to report a mistake finds out about problems while they are small. Add the rule to induction for every new starter and every paralegal or contractor who touches files.

Checking after 90 days that it worked

Review four things. Hours: compare matter time on chronologies and summaries with the quarter before. Errors: count what reviewers corrected, and where. Incidents: count entries on the report-a-mistake log and what was done. Readiness: pick one closed-tool matter at random and see whether the file note answers the four things the Court may ask: whether AI was used, which tool, how the output was checked and supervised, and how the direction's principles were followed. If the file cannot answer them, the routine is not ready yet, whatever the savings.

If you would like help choosing between a closed legal tool and an assistant for your practice, or drafting the staff rule, that is what my AI implementation consultation covers.

Questions family law practices ask about AI

Can I put a client's financial disclosure into ChatGPT?

Not on a free or personal plan. The Family Court's practice direction tells court users to keep confidential or sensitive information out of public AI tools, and material produced under disclosure orders may go into an AI tool only if strict closed-environment conditions are met. A closed tool under a contract that bars training and limits access is the minimum, and you still check every figure.

Do I have to tell the Family Court that I used AI?

No general statement is required, except from expert witnesses, who must certify they have read the direction. Anyone may be asked whether AI was used, which tool, how the output was checked and supervised, and how the principles were met, so keep a short file note on every matter and the answer is ready.

Is AI safe for family violence or child-safety assessments?

No. Those judgements stay with a qualified lawyer. The practice direction says court users must not use AI where there is a risk it will compromise the safety of any court user or child. AI can help with admin around a matter, but risk assessment and advice on safety are not tasks to hand to a tool.

What should I do if a paralegal has already used a free chatbot on a file?

Stop it today, record what was entered and when, assess whether sensitive or protected material was involved, and take advice on any reporting duty. Then put the approved-tool rule in writing and train the team. The practice direction holds court users accountable for the tools they choose, so supervision matters as much as the error.

Further reads

• Family Law Client Intake: An AI Questionnaire and Urgency Triage — An intake questionnaire with urgency triage.
• AI Receptionist for Family Law Firms: Handling Callers in Crisis — Handling callers in crisis without a tool deciding safety.
• Family Law Retainer Agreements: AI Drafts, Lawyer Sign-Off — Retainer drafts with lawyer sign-off.
• Family Law Document Automation: Gavel Workflows or Generative AI? — Document automation against generative AI for forms.
• AI Tools for Lawyers in Australia: A Safe Starting Checklist — A short starting checklist for any Australian firm.
• Which AI Is Best for Lawyers in Australia? Choose by Task — A decision rule for choosing AI by task for an Australian law practice, comparing general assistants with the AI built into practice software.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Federal Circuit and Family Court of Australia, Practice Direction: Use of Artificial Intelligence (PD-AI, issued 29 May 2026) and the Court's release notice; published commentary on earlier AI cases in the Court; OAIC guidance on commercially available AI products; Smokeball and Microsoft Australia pricing pages (read November 2026); vendor pricing for Claude Team.