A firm signs up for an AI product. The partner gives it a run. The output looks confident, well-structured, even well-cited. Then someone asks the question that matters on the file: "Is this right, under Australian law, on these specific facts?" The silence that follows is the real story behind the adoption numbers.
Lawyers Weekly reported this week on analysis from LEAP putting 16% of Australian firms using legal-specific AI in core workflows, against 49% globally. Clio's figure is more generous at 39%, but even that number tells you most firms that have bought into the category haven't committed to it. InfoTrack's general manager of property and financial services, Ajay Kumar, offered a diagnosis worth taking seriously: tools built for global enterprise legal teams "do not always reflect the way Australian lawyers and conveyancers work on local matters."
That observation deserves more attention than it's received.
The explanation that circulates most readily for the Australian adoption lag is a skills problem. Lawyers need more training. Firms need change management. The profession is conservative by disposition and slow to move. There's something to all of that. But it sidesteps the structural issue Kumar is pointing at.
When a practitioner in Sydney runs a query about a conveyancing matter or an employment dispute, what they need is accurate, traceable authority in Australian law, specific to how courts have applied those provisions in the jurisdictions that matter for the file in front of them. A globally trained model, however capable on aggregate, isn't built to deliver that with any reliability.
Global tools reflect global markets. The training data captures what legal information looks like at scale across many jurisdictions. The outputs can be fluent and thoroughly wrong on an Australian matter, wrong about what the legislation says, wrong about how the courts have applied it, wrong about which authority carries weight in the relevant state. A practitioner using such a tool on a live file faces professional responsibility consequences they can see coming. So they pull back. They go back to the manual search. They treat the AI as something to experiment with, not something to build their workflow around.
That professional responsibility exposure is specific, not abstract. A solicitor who passes along advice based on an AI output that misstates the legislative position, or cites an authority that doesn't support the proposition, has an answer to give the Law Society and their insurer. The duty of competence doesn't bend for tools the practitioner failed to verify. Most practitioners grasp this intuitively, which is precisely why they stop short of full reliance.
This is why the adoption curve looks the way it does. People signed up. They tried things. They found the tool didn't answer the question they were asking. Rational behaviour, given the tools on offer.
The concerns practitioners raise most often, according to the survey data, are fabricated outputs and data leakage. These are worth examining closely, because they map directly onto the structural problem Kumar identified.
Fabricated outputs are a function of how most AI systems work: the model predicts what a plausible response looks like, and in legal work, a plausible-looking citation can be completely false. A confident-sounding reference to a High Court decision that doesn't exist, or that held the opposite of what the output claims, is far more dangerous than an obvious gap in the research. It creates the appearance of authority while providing none.
The deeper catch is verification. If a practitioner must check every citation manually before relying on it, they have reconstructed the original research problem with an extra step in front of it. The time saving evaporates. Worse, the practitioner now carries the cognitive load of both the AI output and the verification pass, rather than having done the research once with confidence. Tools that produce unverifiable output don't compress the workflow; they fork it.
Data leakage is a function of infrastructure. Where does the information that a practitioner types go, and who can access it? In a field where client confidentiality is foundational, uploading matter materials to a general-purpose AI service without understanding its data handling is a risk most firms are right to decline.
Both concerns converge on the same underlying problem. A tool built for global enterprise markets, with no grounding in Australian primary sources and no designed transparency around its reasoning, asks practitioners to trust outputs they cannot verify. A practitioner declining to rely on unverifiable output is exercising professional judgment, not revealing a skills gap.
Kumar's diagnosis points toward the fix. The problem sits between what global enterprise tools were designed for and what Australian lawyers and conveyancers do on live files. Closing it requires AI grounded in Australian legal sources, built around how Australian practitioners work, with clear lines on where professional judgment stays in charge.
That means something specific, and we think it's worth being precise about it. It means search that runs against a verified, closed corpus of Australian law, not a general-purpose index that happened to include some Australian materials. It means citations that trace back to their source, so a practitioner can check the authority before they rely on it. It means being able to reason over the firm's own matter materials without those materials leaving a secure environment. And it means being honest about what the tool cannot do, which is at least as important as being clear about what it can.
Australian law is also more jurisdictionally complex than the global adoption conversation tends to acknowledge. The interplay between federal legislation and state court decisions on common law questions, the divergent lines of authority that have developed in commercial and employment matters across different state supreme courts, the particular character of how the High Court's reasoning gets applied at the intermediate appellate level: these are not edge cases. They are the everyday texture of Australian legal practice. A tool that flattens this into generic common-law output is unhelpful precisely when the file is most contested.
More education about how large language models function will not help a Sydney conveyancer trust an output that may or may not reflect how the Torrens system operates in New South Wales. The answer is tools built for the specific shape of Australian legal practice.
This is the territory Habeas was built for. The 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 generated from pattern-matching across an opaque global corpus. Foundational research that used to take a full morning can be completed in minutes, with every citation traceable to its source document.
We are not claiming this solves the adoption problem across the whole market. Plenty of firms face genuine barriers that a better-designed product does not remove on its own. But the practitioners we work with, from barristers running brief-review workflows to GCs managing employment, privacy, and contracts across a wide surface area, adopt Habeas precisely because the Australian-law grounding is verifiable. They can follow the citation. They can check the reasoning. They can decide whether to rely on it.
That's what makes the difference between a tool people experiment with and a tool people build their practice around.
The 16% figure is a rational response to a category that, for most of its existence, has asked Australian practitioners to trust tools that weren't designed for their jurisdiction or their workflow. InfoTrack's diagnosis should be read as a market observation with consequences: the firms and platforms that close the adoption-to-capability gap will do it by meeting practitioners where they work, on Australian files, with outputs that can be verified.
We think that's a more interesting and more honest framing of the opportunity than the standard adoption narrative, which treats the profession as the problem to be overcome. The tools were the constraint. That, at least, is solvable, for the firms willing to be precise about what solving it requires.
If you want to see what Australian-first, citation-grounded legal research looks like, book a demo 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.
