The question arrives at 4pm on a Thursday. A senior stakeholder needs a view on whether a new commercial arrangement sits within the bounds of the Competition and Consumer Act before the board papers circulate on Friday. They are not asking for a memo. They want a view, confidently expressed, by morning.
So the GC opens the practice note they last used six months ago, and starts reading.
This is where the problem begins. The memo you build from a practice note is only as current as the practice note, and neither of you will know when it stopped being true. Parliament amends a provision; an appellate court reframes the test. The updated position takes time to filter into secondary literature. A GC advising on whether a commercial arrangement engages the competition provisions of the CCA might be working from commentary written before a significant legislative change to the effects test, or from external advice predating a subsequent Full Federal Court decision. The primary law has moved; the resource in the GC's hands has not. The advice is not fabricated. It is just wrong, in a way that will show clearly when someone asks for the source.
The GC knows this. They also know that calling external counsel at 4pm on Thursday, for a view by Friday morning, produces a different problem: delay, cost, and the awkward reality that the business will have moved on before the memo arrives. So they try something else. They type the question into one of the large language model tools they have been using for other tasks.
The response sounds authoritative. There are citations. There is structure. There is a confident summary of the applicable test. The GC reads it twice, and on the third read they notice a case name that looks slightly wrong. They search for it. It does not exist. They search for a variant. Nothing. The citation is fluent and it is fabricated, and had they not checked, it would have landed in documented advice.
This is the specific risk that generic AI tools introduce into legal workflows, and it is worse than slow research because it is invisible. A gap in a research note is obvious. A confident, wrong citation looks like research. The professional standards guidance on AI use in legal practice is unambiguous: verification sits with the practitioner. A tool that generates plausible-sounding citations expands the verification burden precisely because the starting point cannot be trusted.
The GC closes the tab and opens Habeas instead.
The semantic Search Engine runs a natural-language query across more than 300,000 Australian cases and pieces of legislation. The question about the commercial arrangement and the CCA does not need to be reconstructed into Boolean syntax. It is asked in the same terms the business posed it. The results come back from a closed dataset of primary Australian legal sources, with citations the GC can follow to their source. Each one resolves. Each one says what it is said to say.
At this point in the Thursday evening, the GC is not fighting their research tool. They are doing the actual work: reading the authorities, tracing how the relevant test has been applied, identifying where the arrangement sits. The unfair contract terms reforms that commenced in November 2023 are relevant to a related question about the standard-form agreement the new arrangement involves. The regime changed, civil penalties were introduced, the threshold defining which business contracts fall within scope shifted. Habeas returns the current position. The GC follows the citations, confirms the holding, and documents the verification. The research trail is auditable from the start.
This is the line that separates usable AI research from the kind that creates risk. One GC put it plainly: "The Australian-law focus and the depth of nuance in the answers is a major differentiator. It materially changes my confidence and speed when forming legal views." Australian law is not American law with the names changed. The statutory architecture is different, the court hierarchy is different, and jurisdiction-specific nuance in areas like consumer protection, employment, and privacy requires precision that an undifferentiated training corpus, weighted heavily toward American and UK volume, cannot reliably deliver.
By 10pm the primary law question is answered. The GC has a clear view on the CCA, grounded in Australian authorities with a traceable citation for each proposition.
Then they turn to the matter itself, and this is where the limit sits. Habeas searches a defined corpus of public Australian law. The matter documents, the internal correspondence, the draft commercial agreement, sit outside that corpus. The primary law research is done, but applying it to the specific facts requires the GC's own materials. A separate Habeas capability, Document Stores, handles that layer: upload the matter documents, query across them alongside the public law, receive answers grounded in both. The GC already has the verified authorities in one hand; Document Stores lets them hold the matter in the other.
The board paper exists by Friday morning. Every proposition rests on a traceable citation. When someone asks "where did that come from?" the answer is a real case, a real provision, a source that can be opened and read by anyone in the room.
In regulated sectors this is not an abstract comfort. ASIC enforcement actions and ACCC investigations can reach the materials a GC used to form a legal view, including the research process behind documented advice. A research trail grounded in verifiable Australian primary sources looks substantially different from one supported by a generic AI output that cannot be independently confirmed. The practitioner whose research began with a verified corpus is in a materially better position when that question arrives.
Speed and defensibility, when the research is grounded in verified Australian primary law from the start, become the same exercise.
If that Thursday scenario is familiar, see how the semantic Search Engine works 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.
