Legal AI Ethics in Australia: What the Law Society Says
Abstract — Legal AI ethics Australia questions do not turn on new rules. Every state law society and the Law Council of Australia has approached generative AI the same way: the existing conduct obligations already cover it. Competence, confidentiality, supervision, honesty and the paramount duty to the court apply to work produced with AI exactly as they apply to work produced without it. What has been added is court practice guidance and an expectation that you verify. Here is the consolidated position.
Each Australian law society publishes its own guidance separately, which means practitioners across jurisdictions have to assemble the picture themselves. This page consolidates it. The consistent message on legal AI ethics Australia wide is unglamorous and important: nothing about your duties has changed, and that is precisely why the risk is real, because a fabricated citation is a breach of an obligation you already had.
The one sentence version. You may use AI. You remain fully responsible for everything produced with it, and you cannot delegate the duty to verify.
What State Law Societies Have Said About AI
Guidance is issued and updated at different times by different bodies, so confirm the current version with your own regulator before relying on any summary, including this one. The themes below are consistent across all of them.
The Law Council of Australia has addressed AI use at the national level, framing it through existing professional obligations rather than proposing a separate regime. The Australian Solicitors' Conduct Rules, adopted in most jurisdictions, remain the operative source.
The Law Society of New South Wales has published guidance for solicitors on generative AI, emphasising competence, confidentiality and verification. New South Wales practitioners also work under the most detailed court guidance in the country, since the Supreme Court of New South Wales practice note restricting generative AI in affidavits and expert reports applies to work filed there.
The Law Institute of Victoria has issued guidance to Victorian practitioners covering similar ground, with attention to client confidentiality and the risks of putting client material into third party systems.
The Queensland Law Society has published material for Queensland practitioners on the ethical use of AI, again grounded in the existing conduct rules rather than in new obligations.
The Law Society of Western Australia, the Law Society of South Australia, and the societies in Tasmania, the ACT and the Northern Territory have each addressed AI use for their members, with the same underlying framework.
Courts have gone further than the societies in one respect: several have issued binding practice requirements. Because these differ between courts, a practitioner appearing in more than one cannot assume a single standard applies.
The practical consequence of this structure is that your obligations come from three places at once: your conduct rules, your law society's guidance, and the practice notes of every court you appear in.
5 Ethical Obligations When Using AI in Practice
Competence. You must understand the tool well enough to know its limits. Using a system whose outputs you cannot evaluate is a competence problem, not a technology problem. In practice this means knowing whether the tool retrieves from real sources or generates from memory, because that single fact determines how much verification the output needs.
Confidentiality. Client information must be protected. Entering it into a third party system is a disclosure. Before doing so, resolve the vendor's position on storage location, retention, training use and deletion, and consider whether the client's instructions permit it. Client information is also personal information under the Privacy Act 1988 (Cth), and cross border disclosure engages Australian Privacy Principle 8.
Supervision. You are responsible for work produced by those you supervise, including work they produced with AI. The workable control is requiring the verification step to be performed and recorded. It has a useful side effect: verification teaches juniors the judgment that unsupervised AI use erodes.
Honesty and the duty to the court. The paramount duty means an authority you put before a court must exist and must support what you cite it for. This is where the profession's exposure is sharpest, and it is not a new rule.
Accuracy and diligence. Every citation opened, every section read in the version in force, nothing filed or sent unverified. "The software produced it" has never been a defence to anything.
A sixth, sitting alongside: costs disclosure. If AI materially reduces the time a matter takes, your costs estimates and any fixed fee should reflect it. Charging for time not spent is a costs assessment problem regardless of how the efficiency arose.
Do You Have to Disclose AI Use to Clients?
The short answer is that there is no blanket obligation to disclose ordinary use of AI tools, and there are several situations where disclosure is required or plainly advisable.
Where a court requires it. Some practice notes require disclosure or certification about AI use in particular documents. This is binding, and it varies by court, so check the notes for every court you appear in.
Where the client has asked, or has a policy. Increasingly, corporate clients include AI use terms in their outside counsel guidelines. Those terms are contractual, and they bind you.
Where confidential material is involved. Uploading client material to a third party system is a disclosure decision the client is entitled to be part of. Many practitioners have settled on obtaining consent for anything beyond public material, which is a sensible default.
Where it affects the fee. If efficiency changes the estimate, that engages your costs disclosure obligations regardless of the cause.
Where the client would be surprised. The practical test most practitioners apply, and a good one. If a client would be unhappy to learn afterwards how something was produced, tell them beforehand.
Where it does not arise. Using a research tool to find an authority you then read yourself is not different in kind from using a subscription database. Nobody discloses which search engine they used.
Consequences of Getting It Wrong
The consequences are real and they are already being applied, in Australia and overseas.
Professional discipline. Practitioners have been referred to regulators over material containing citations that could not be found. The breach is not "using AI"; it is filing something unverified, which engages competence, diligence and the duty to the court simultaneously.
Judicial criticism in open court. Reasons for judgment are published. Criticism of a practitioner for putting non-existent authority before a court is permanent and searchable, and the reputational cost typically exceeds any formal sanction.
Costs consequences. Courts can and do make costs orders where time has been wasted.
Client loss and insurance exposure. A client who discovers their confidential material was uploaded to a system without consent has a complaint. A negligence claim arising from unverified output is exactly the kind of claim professional indemnity insurers examine closely.
The pattern in every reported instance is the same, and it is worth stating plainly: the failure was never the tool. It was a person filing something they had not checked. The control that prevents all of it takes minutes per authority on AustLII.
The four situations where practitioners actually get caught
Naming the circumstances is more useful than naming the rules, because the rules are already known and the circumstances are where they fail.
Time pressure the night before. The single most common context. A document due in the morning, an authority that looks right, and a verification step that gets skipped exactly once. The control is a rule written down in advance, because a rule you rely on judgment to apply will not survive the night it matters.
Delegation without a verification instruction. A junior or paralegal asked to "find me a case on this" and a supervisor who assumes checking was implied. Make it explicit, and require it to be recorded.
Working outside your usual area. A practitioner briefed into unfamiliar law has less capacity to notice that an answer is subtly wrong. This is precisely where AI is most useful and where its output most needs checking, which is an uncomfortable combination.
Trusting a tool that has been right many times. Calibration drifts with familiarity. A hundred correct answers do not change the probability that the next one contains a paraphrase that overstates the section, and confidence built on a track record is the wrong kind of confidence here.
None of these are technology failures, and none are addressed by choosing a better vendor. They are addressed by a rule, written before it is needed, that says nothing goes out unverified and names who is responsible.
Ask.Legal's Approach to Verification
Ask.Legal is built to make the verification obligation this guide describes easy rather than onerous: it aims to retrieve from Australian legislation and case law before answering, name the jurisdiction on state law topics, and attach citations that open in a click, so the check that protects a practitioner's duty to the court takes seconds rather than minutes. See how it's built or try it free. For the conduct rules and court guidance referenced throughout, see the Law Council of Australia's AI portal and, for verification itself, AustLII.
Frequently Asked Questions
Is using AI ethical for Australian solicitors? Yes, subject to existing obligations. Law societies have addressed it through the Australian Solicitors' Conduct Rules rather than through new rules.
Do I have to tell clients I used AI? Not as a blanket rule. You must where a court practice note or the client's own terms require it, and it is advisable where confidential material or the fee is affected.
Can I upload client documents to an AI tool? Only after resolving retention, training use and storage location, and considering client consent and privilege.
What happens if AI produces a fake citation and I file it? You may face regulatory referral, judicial criticism and costs consequences. Responsibility sits with you, not the tool.
Which rules actually apply? Your conduct rules, your law society's guidance, and the practice notes of every court you appear in, together.
Key Takeaways
No Australian law society has created new AI rules; all have applied the existing conduct obligations.
Competence, confidentiality, supervision, honesty and diligence cover the field between them.
Disclosure is required where a court, a client's terms or the fee position demands it, not as a blanket rule.
Every reported failure has been a person filing unverified material, which is preventable in minutes.
Sources
Australian Solicitors' Conduct Rules, as adopted in most Australian jurisdictions
Law Council of Australia, and guidance published by the Law Society of New South Wales, Law Institute of Victoria, Queensland Law Society and other state and territory law societies
Supreme Court of New South Wales practice note on generative artificial intelligence, and equivalent guidance from other Australian courts
Privacy Act 1988 (Cth) and the Australian Privacy Principles, including APP 8
State and territory legal profession costs disclosure requirements
Ask.Legal is built to make the verification obligation easy rather than onerous: retrieval from Australian legislation and case law, the jurisdiction named on state law topics, and citations that open in a click so the check that protects you takes seconds.
See how Ask.Legal is built to support ethical AI use in practice
This article is general information about the law of Australia as at 2026, not legal advice. For advice on your circumstances, consult a qualified Australian legal practitioner.