Artificial intelligence has moved out of the "coming soon" column of legal industry predictions and into the software your business already pays for. Tools like ChatGPT, Microsoft Copilot and Claude now draft emails, summarise contracts and answer plain-English questions, and legal-specific products such as Harvey are marketed directly at law firms. For an Australian business owner, the practical question is no longer whether AI will touch legal work. It is what AI can safely be used for, where it gets things wrong, and who is accountable when it does.
What kind of AI is actually doing legal work
Artificial intelligence is a broad label for machines performing tasks that would normally require human intelligence. Under that umbrella sit two very different families of technology, and only one of them is driving most of the current change.
Machine learning
Machine learning describes systems that improve at a task by learning from data rather than being explicitly programmed for every step. It powers analytical tools that classify, rank and predict: e-discovery software that sorts thousands of documents for relevance, contract review tools that flag unusual clauses, and platforms that assess litigation risk.
Generative AI
Generative AI creates new content. As the Federal Court of Australia defines it in its Use of Generative Artificial Intelligence Practice Note, generative AI produces text, images, music, audio and video based on a user's prompts. This is the technology behind ChatGPT, Claude, Google Gemini, Microsoft Copilot and Harvey. Where analytical AI reads what exists, generative AI writes something that did not exist before, and that is what makes it powerful and, as discussed below, dangerous in legal work.
Where AI is already used in legal work
The Federal Court's practice note observes that generative AI can increase efficiency in litigation, reduce legal costs and improve access to justice, and the same logic applies to everyday commercial legal work. Current uses include:
- Legal research: AI tools summarise cases and statutes, locate authorities and answer questions framed in everyday language, cutting down hours of database searching.
- Contract work: businesses and their lawyers use AI to summarise long agreements, compare versions, extract key dates and obligations, and produce first drafts.
- Client intake and chatbots: automated assistants take initial enquiries around the clock, gather the basic facts of a problem and direct the enquiry to the right person.
- Document review in disputes: machine learning sorts and prioritises discovery material far faster than a team reading documents one by one.
- Predictive analytics: statistical tools analyse past decisions and outcomes to inform strategy, settlement positions and cost estimates.
Each of these is genuinely useful, and each assumes a competent person will check the output before it is acted on. That assumption is the entire risk profile in one sentence.
The hallucination problem: when AI invents the law
Generative AI does not retrieve information the way a search engine does. It predicts the next most plausible word, which means it can produce output that is confident, fluent and completely wrong. The Federal Court's practice note is blunt about it: generative AI may give users fictitious cases, citations and quotes, references to legal sources that do not exist, incorrect or misleading statements of the law, and factual errors, and it may confirm that incorrect information is accurate if asked.
The consequences of treating that output as reliable are now documented in case law on both sides of the Pacific. In the United States, the Avianca matter produced the same lesson: in 2023 lawyers were sanctioned after filing court submissions citing six non-existent cases generated by ChatGPT, with the court observing that technology does not remove a lawyer's responsibility to ensure the accuracy of filings. Closer to home, in Pasuengos v Minister for Immigration and Citizenship (No 2) [2026] FedCFamC2G 96, a practitioner was referred to the South Australian Legal Profession Conduct Commissioner after submissions filed in the Federal Circuit and Family Court contained case citations that an AI tool had fabricated and that no one had verified before filing.
The Federal Court has said plainly that presenting false or inaccurate information to the court is unacceptable, and that courts may order disclosure of how generative AI was used, with adverse costs orders and professional consequences available where it is misused. Every citation and every statement of the law produced by an AI tool needs to be checked against a reliable source before it goes to a client, a regulator or a court.
The rules that already apply to AI in Australia
Australia does not yet have a dedicated AI statute. The Commonwealth's Voluntary AI Safety Standard, published in September 2024 and updated in December 2025, sets out guardrails for safe and responsible use of AI, and from October 2025 a simpler "Guidance for AI Adoption" replaced its ten guardrails with six essential practices. The government has been explicit that the standard is voluntary and does not create new legal obligations. Mandatory guardrails for high-risk AI have been proposed and consulted on, but they are not law, so the rules that actually bind you are the ones that already exist:
- Privacy: If you feed personal information into an AI tool, the Privacy Act 1988 (Cth) applies. Australian Privacy Principle 11 requires an entity holding personal information to take reasonable steps to protect it from misuse, interference and loss and from unauthorised access, modification or disclosure. Information entered into a generally accessible generative AI tool may become available to other people, and you may not know where it is stored or who can access it. Client names, employee details, commercial terms and anything subject to legal professional privilege should not be pasted into a public tool without assessing that risk first.
- Professional conduct: Solicitors' duties of competence and diligence, and the paramount duty to the court, apply with full force to AI-assisted work. The Law Society of New South Wales published a guide to responsible use of AI in January 2026, and the Queensland Law Society's Guidance Statement No 37 sets out a framework for ethical use of AI in legal practice. Using AI to produce work you have not checked can amount to unsatisfactory professional conduct, and the Legal Profession Uniform Law framework treats such conduct seriously.
- Courts: The NSW Supreme Court Practice Note SC Gen 23 has applied to all proceedings since 3 February 2025, and the Federal Court's practice note has applied since April 2026. Both expect anyone using generative AI in litigation to understand its limitations, and both permit courts to require disclosure of how it was used.
- Intellectual property: In Commissioner of Patents v Thaler [2022] FCAFC 62, the Full Federal Court held that an AI system cannot be an inventor under the Patents Act 1990 (Cth): an inventor must be a natural person. Ownership of AI-generated content, and the terms on which AI vendors claim rights in your inputs and outputs, are live issues in any contract with an AI provider.
Existing laws on misleading conduct, confidentiality and contract also continue to apply to AI use. The absence of an AI-specific statute does not mean there is a gap; it means the ordinary law does the work, case by case.
When you need a lawyer for your AI questions
The judgement calls in this area are exactly the kind a lawyer should help with, because the cost of getting them wrong can be a conduct complaint, a breached confidentiality obligation or an adverse costs order. An Artificer Legal lawyer can help you:
- review an AI vendor's terms to see who owns the outputs, what rights the vendor takes in your data, and what indemnities and warranties protect you;
- draft an AI use policy for your staff covering what can and cannot be entered into these tools;
- assess whether a proposed use of AI complies with the Privacy Act before client or employee data goes anywhere near it;
- work out who owns AI-assisted content and what protection is available for it; and
- respond if AI-generated material has already been filed, published or relied on and a regulator, court or counterparty is asking questions.
The duty that AI has not changed
The single point worth holding onto is that AI changes the risk, not the accountability. Whatever produced the document, the person who files it, sends it or relies on it remains responsible for it. The lawyer who submits a hallucinated citation, and the business that acts on AI-generated advice without checking it, face the same consequences as if a human had produced the error. The gatekeeping duty is unchanged: verify the authorities, protect client information, and keep a human accountable for every decision.
To summarise: generative AI is now genuinely useful for legal research, contracts and intake, but it fabricates authorities and confirms its own mistakes. Australia regulates it through existing law, especially the Privacy Act and professional conduct rules, plus court practice notes that require disclosure and verification. The voluntary national standard sets out sensible governance practices but is not a compliance shortcut. Before you rely on AI output, or let your staff feed client information into it, get advice on the specific use, because the responsibility sits with you either way.