Can lawyers use AI for file notes in Australia?

The short answer, and the four conditions

The Australian guidance is consistent. AI is a tool, and the solicitor stays responsible for the work and the client's information. For file notes that means four conditions.

1. Confidentiality. Client information only goes into a tool that contractually keeps it confidential, does not train on it, and is not a public chatbot.
2. Privilege. The way the tool handles the information must be consistent with keeping it confidential, because privilege depends on confidentiality.
3. Accuracy and supervision. A person who knows the matter reads the draft against what was actually said, and fixes it, before it becomes the record.
4. Policy and transparency. The firm has an AI policy naming the approved tools, and can tell a client how AI is used if asked.

ASCR r 9: confidentiality is the main gate

Rule 9.1 of the Australian Solicitors' Conduct Rules says a solicitor must not disclose information that is confidential to a client and acquired during the engagement to anyone outside the practice. There are two carve-outs that matter for software. Rule 9.1.2 allows disclosure to a person 'otherwise engaged by' the law practice for the purposes of delivering or administering legal services in relation to the client. Rule 9.2.1 allows disclosure the client expressly or impliedly authorises.

Firms already rely on this footing for cloud practice management, email hosting and outsourced typing. A vendor engaged under confidentiality terms is in a very different position to a public tool whose terms let it keep and reuse what you type in. Our longer piece on ASCR file note obligations covers the rules that sit around the note itself.

The Law Society of NSW's 'A Solicitor's Guide to Responsible Use of Artificial Intelligence' puts the risk bluntly. It treats copying confidential client information into a public generative AI system as akin to putting it into the public domain, which could breach r 9 and cost the client their privilege. The guide names five rules to keep front of mind: rr 4, 9, 17, 19 and 37.

The joint regulator statement (6 December 2024)

On 6 December 2024 the Law Society of New South Wales, the Legal Practice Board of Western Australia and the Victorian Legal Services Board and Commissioner published a joint 'Statement on the use of artificial intelligence in Australian legal practice'. These are the regulators in the three Uniform Law states. The statement does not ban AI. It sets expectations:

• Confidentiality. Lawyers cannot safely enter confidential, sensitive or privileged client information into public AI chatbots or copilots like ChatGPT, or any other public tools. For commercial AI tools used with client data, lawyers need to carefully review the contractual terms to make sure the information will be kept secure.
• Independent judgement. AI cannot reason, understand or advise. The lawyer exercises the forensic judgement.
• Competence and accuracy. Lawyers must check that AI output is accurate and not misleading to clients, courts or other parties.
• Risk-based policies. Firms should have clear AI policies covering which tools are approved, who can use them, for what, how junior staff are supervised and how output is verified.
• Lower-risk tasks first. Keep AI to lower-risk, easily verified tasks. A summary of a call the solicitor was on is one of them.
• Transparency. Be able to tell clients, and courts where necessary, when and how AI was used and what it meant for costs.

What the state bodies add

Queensland Law Society - Guidance Statement No. 37

QLS Guidance Statement No. 37, 'Artificial Intelligence in Legal Practice' (31 May 2024), is the most practical of the documents on vendor checks. It tells practitioners to take reasonable steps to ensure any AI tool does not misuse access to confidential data or unduly expose it, and it lists the questions to ask: where is the data processed, how and where is it stored, who has access, does the provider keep what you submit, and does it claim any rights over it. Read both the terms of service and the privacy statement, and be very hesitant if they do not answer those questions clearly.

On supervision under r 37, QLS says each piece of work produced with an AI system should be checked by someone who knows the area of law well enough to spot errors, and that AI tools should only be used under a clear firm policy. On disclosure, the more significant the AI contribution to the client's work, the stronger the obligation to tell the client, and QLS suggests covering AI tools in the costs agreement or engagement letter.

Law Institute of Victoria - 2025 ethics guidelines

The LIV published 'Ethical and Responsible Use of Artificial Intelligence' guidelines in 2025. The LIV goes a step further on consent: it tells practitioners never to put confidential or privileged information into open-source or commercial AI systems without client consent, and to review a tool's privacy policy and terms of use and engage with the vendor before adopting it. In Victoria, the safe course is consent through your engagement terms.

Privilege and third-party processors

Privilege protects confidential communications. Lose the confidentiality and you lose the footing for the claim. The High Court made the point in Glencore International AG v Commissioner of Taxation [2019] HCA 26: privilege is an immunity from compelled disclosure, a shield, and it does not restore confidentiality to information that has already lost it. QLS cites Glencore and asks the question that matters: are the tool's terms of service incompatible with an intention to keep the material confidential?

Waiver at common law turns on whether the privilege holder's conduct is inconsistent with maintaining confidentiality (Mann v Carnell (1999) 201 CLR 1). Giving an ATT note to a provider contractually bound to keep it confidential is the same kind of limited disclosure firms make every day to cloud hosts and typists. Pasting it into a consumer tool whose terms allow retention, review or training is much harder to defend. We are not aware of an Australian decision that squarely decides the AI-vendor point, so treat this as a risk to be managed through contract terms, not a settled question. See are attendance notes privileged? for how privilege attaches to the note in the first place.

Court practice notes on generative AI

The court documents are aimed at proceedings, not internal file notes. They still matter, because an attendance note can end up feeding an affidavit.

NSW Supreme Court Practice Note SC Gen 23, 'Use of Generative Artificial Intelligence (Gen AI)' was first issued on 21 November 2024, reissued on 28 January 2025 and commenced on 3 February 2025. The key points for note-taking:

• Transcription is carved out. Paragraph 6 says Gen AI, for the purposes of the practice note, does not include technology that merely corrects spelling or grammar, provides transcription or translation, or assists with formatting without generating substantive content. A tool that writes a structured note does generate content, so assume the practice note applies where the material relates to proceedings.
• Summarising transcripts is a permitted use. Paragraph 9B lists acceptable uses, including generating chronologies and 'the summarising or review of documents and transcripts'.
• Restricted material. Under paragraph 9A, information subject to suppression or non-publication orders, the implied (Harman) undertaking, material produced on subpoena, or a statutory prohibition on publication must not be entered into any Gen AI program unless you are satisfied it stays within a controlled environment, the supplier is bound by confidentiality restrictions, it is used only for that proceeding, and it is not used to train any model.
• Evidence. Paragraphs 10 to 13 say Gen AI must not be used to generate the content of affidavits, witness statements or character references, or to alter or rephrase a witness's evidence. Preparatory work is allowed. An affidavit must disclose that Gen AI was not used to generate its content.

So if an AI-drafted note later supports an affidavit about a conversation, the evidence must still be the deponent's own recollection in their own words. The note can refresh memory. It cannot be pasted in.

Victoria now has a practice note as well: the Supreme Court of Victoria's Practice Note SC Gen 25, 'Use of Artificial Intelligence by Court Users', was issued and commenced on 14 May 2026, replacing the Court's 'Guidelines for litigants: responsible use of artificial intelligence in litigation' of May 2024. Other courts and tribunals have issued their own documents, so check the ones you appear in.

Accuracy and supervision: the note is still yours

Rule 4.1.3 requires legal services to be delivered competently and diligently. Rule 37 makes the supervising solicitor responsible for work done under them. An AI draft changes neither. Build three habits into the workflow:

• Review the same day. Read the draft while the call is fresh. Check names, dates, amounts, deadlines and limitation periods first.
• Check the instructions and advice line by line. The instructions received and the advice given are what the note exists to prove. Make sure a qualification you gave ('subject to seeing the contract') has not been flattened into an unqualified statement.
• Lock the final version. A note that can be edited quietly months later is weaker evidence. Keep the reviewed version fixed, with any later correction added as a dated amendment. See file notes as evidence.

A reviewed note looks like this. The client and matter are invented.

If you want a starting structure, the attendance note generator and the law firm file note compliance checklist are free to use.

Recording is a separate legal question

Many AI note-takers work by recording the call or joining the meeting as a bot. That raises listening and surveillance devices legislation before you ever get to the conduct rules. NSW, WA, SA, Tasmania and the ACT generally require all parties to consent to a private conversation being recorded. Victoria, Queensland and the NT allow a party to the conversation to record it, with limits on what can be done with the recording. The Telecommunications (Interception and Access) Act 1979 (Cth) separately prohibits intercepting a communication passing over a telecommunications system. See recording phone calls in Australia for the state detail, do AI note-takers record your calls? for which tools capture audio, and meeting bots vs transcript-only tools for the trade-offs.

Data residency and the Privacy Act

No conduct rule says client data must stay in Australia. The QLS guidance makes 'where is the data being processed and stored' a required question, and there are good reasons to prefer an Australian answer. Data held offshore can be reached by foreign legal process. If the Privacy Act 1988 (Cth) applies to your practice, APP 8 and s 16C make you accountable for what an overseas recipient does with personal information you disclose to it, and APP 11 requires reasonable steps to secure what you hold. The OAIC's 'Guidance on privacy and the use of commercially available AI products' (October 2024) recommends not entering personal information into publicly available generative AI tools at all.

Many small practices fall under the Privacy Act's small business exemption. Rule 9 has no exemption, so the vendor questions are the same either way.

Vendor due-diligence checklist

Put these to any AI note vendor in writing and keep the answers with your AI policy.

Our own answers: CallNote never records calls, never joins a meeting as a bot and never stores call audio. It works from a transcript that already exists: pasted, uploaded as a .vtt or .txt file, forwarded by email, or received automatically from Dialpad or Aircall. Data is hosted in Sydney (AWS ap-southeast-2), encrypted, and not used to train AI. You review every draft, then publish it (SHA-256 checksum, append-only amendments, audit log, PDF export). There is no native integration with LEAP, Smokeball or Actionstep, so the note reaches the matter by PDF or copy and paste. More on how CallNote works for law firms.

Put it in a one-page AI policy

• Name the approved tools. Public chatbots are prohibited for client information.
• File the vendor due diligence with the date, and re-check it yearly.
• Say who reviews drafts prepared by junior staff and paralegals, and require same-day sign-off.
• Exclude restricted material unless SC Gen 23 para 9A or the local equivalent is satisfied.
• Add the technology clause to the CA, and bill for the time actually spent.

Common questions