On 23 June, Australasian Lawyer ran a headline that travelled well beyond the legal press: an AI-only law firm had won a contested trial, a claimed world first. Garfield, a UK practice built around AI systems rather than admitted solicitors, had secured a verdict in a disputed matter. By early July the Queensland Law Society's Proctor was running its own analysis, examining what the result might mean for Queensland practitioners.
We have been reading the detail. The story is worth telling accurately, because the compressed version that travels fastest is already shaping the wrong conversation.
Garfield produced the documents. A human barrister ran the advocacy. The matter was a £7,000 small claim. That framing changes nothing about what was achieved: building AI systems capable of producing reliable litigation documents is hard, and the Garfield team has done serious work. But it does change the claim. Someone was still accountable, in person, for what was argued in the room. A person had to stand behind the work.
It is also worth noting the procedural environment in which this happened. UK small claims in the County Court are specifically designed to function without lawyers. Documents are the primary mechanism; oral evidence is limited by design. The document layer is relatively more determinative in that setting than in higher-stakes litigation, and relatively less dependent on in-room legal judgment. The result is real. The environment was about as controlled as a contested proceeding can be.
For Australian practitioners, the more consequential story sits in the regulatory gap this result exposes.
Garfield operates under UK rules. No Australian state or territory currently licenses an AI-only practice. A practising certificate here still has to sit behind a human. The question Australian regulators have been deferring, at what point does AI-assisted document preparation become the practice of law, is one a UK firm's trial win has now placed on the table. It will not go quietly back onto the shelf.
The statutes give some shape to the problem without resolving it. The Legal Profession Act 2007 (Qld) defines legal practice to include legal services, which in turn encompasses the preparation of legal documents. The Legal Profession Uniform Law, applying in New South Wales and Victoria, takes a similar approach. Both create offences for engaging in legal practice without a practising certificate. Whether AI systems preparing documents for a fee, without an admitted practitioner's involvement at the drafting stage, falls within those definitions is a question no Australian court or regulator has yet answered on facts like these. The honest position is that the statutes were written against a background where the question could not have been posed in this form.
The QLS Proctor's Queensland-specific analysis reflects a genuine concern: the result lands differently in a jurisdiction where unauthorised practice carries statutory weight and where the Law Society has a formal role in maintaining standards. The document-plus-human-advocate model Garfield used may be distinguishable under current Australian rules. Or it may not be. The honest answer is that nobody knows yet, because the rules were written before the question existed in this form.
This is where we think practitioners should be watching rather than waiting. The document layer of legal work, drafting, reviewing, synthesising, preparing submissions, is being compressed by AI systems faster than any regulatory framework can adapt. The Garfield result is one data point in that compression. The arc runs in one direction.
What is stable, across any regulatory configuration, is the accountability structure. A person has to stand behind the analysis. That person needs to understand what the AI produced, verify it against real sources, and be able to defend it if asked. Systems that generate fluent-sounding output with opaque sourcing become a liability in that world. The verification step is the work, not a formality added on top of it.
The Federal Court's Generative AI Practice Note, published in April 2026, provides the clearest available codification of what that accountability looks like in practice. A practitioner who relies on AI-generated material must personally confirm that cited authorities exist and support the stated proposition, that facts are based on what the party can prove, and that they can account for AI involvement if the Court requires. The barrister who appeared at the Garfield trial was, whether they understood it in those terms or not, fulfilling exactly that function. Australian regulation is moving toward making explicit what a careful lawyer has always had to do; the question is whether the document-preparation layer will ever be permitted to sit outside that accountability chain entirely.
There is also a supervision liability point that deserves attention, and that practitioners operating in firms need to think through. If an Australian firm used externally-prepared AI documents without their own admitted practitioners reviewing the drafting, the professional liability does not redistribute to the AI system. It concentrates on whoever signed the filing. The efficiency gain in preparation does not reduce the professional exposure; it may sharpen it, because the lawyer who files the document now has to account for work they did not generate. The supervision cost does not disappear; it shifts.
The Garfield outcome sharpens that point rather than displacing it. The human barrister who stood in the room was, presumably, in a position to account for every document in the brief. Whether you call that a feature of the Garfield model or a condition of its existence is a question of framing. Either way, the accountability layer did not disappear; it remained human, and it remained essential.
Australian regulators will now have to articulate where the line falls. Bar associations, law societies, and the broader profession will have opinions. The debate will be uneven and slow relative to the pace of development, which is the normal state of affairs when a technology moves faster than the institutions charged with managing it. What practitioners can do now is build their own AI workflows around the accountability requirement, rather than around the hope that the question won't arise.
Habeas was built for exactly that structure. Habeas' Search Engine scans over 300,000 Australian cases and pieces of legislation in seconds, with results grounded in a closed dataset of legitimate Australian legal sources, so they are verifiable and traceable, never hallucinated. Every citation is readable back to its source document. That is the condition under which a person can responsibly stand behind AI-assisted work, whether the regulatory environment looks like today's or tomorrow's.
Research that used to take a full morning can now be completed in minutes. That is the compression the Garfield document layer demonstrated. The difference for Australian practice is that every Habeas output is anchored in Australian primary law, under an accountability framework that fits how Australian practice already works, not imported from a foreign model and retrofitted.
The AI-only firm headline will keep circulating, and the secondary headlines it spawns will get progressively less careful about the detail. The practitioners who read past the summary, who see that AI compressed the document work while a person remained accountable for the argument, are the ones best placed for whatever the regulatory framework becomes. The document layer is most of the hours. Understanding who owns the hours, and on what terms, is now a live question.
See for yourself at habeas.ai.
If you want to try for yourself or get in contact, book a demo with us here. We also offer the capacity for self-serve individuals to sign up, and subscribe or register a free trial at app.habeas.ai.
The legal research in this article was conducted and every citation verified using Habeas, the Australian legal AI research platform.
Hero image: Colin Dean on Pexels
