The submissions arrive at 4:47pm. The hearing is at 10am. Opposing counsel has structured their interlocutory argument around three authorities, one of which you half-recognise from a matter two years ago, in a division you do not practise in regularly. The real questions are whether those authorities still represent the law as stated, whether any subsequent decisions have qualified them, and whether there is a Court of Appeal decision from the last eighteen months that the other side has quietly omitted. You have until morning to find out.
This is where the stress-test actually happens: in the hours after the other side's material lands. Read opposing submissions closely enough and you will find the seams. A primary authority that has not aged well. A proposition stated more broadly than the cases actually support. A line of reasoning that holds together only if a particular recent decision does not exist. The problem has never been knowing those seams are there. The problem is finding them systematically, in an evening window with a hearing at ten.
So she sits with the submissions. She works through the three authorities. She knows the lead case well enough: it is sound. The second she recalls from a seminar paper three years ago, the proposition overstated even then. The third she does not know at all: a single-judge decision from the Federal Court, Western Australian registry, twelve months old, cited confidently for a balance of convenience formulation that carries the plaintiff's whole argument. She does not remember anything from that jurisdiction that would contradict it. Conference draws a blank. She considers running a manual search, but the volume of subsequent authority to trawl before morning is prohibitive.
That is the gap. Memory and conference will find the cases she already knows. They will not reliably surface the Full Court decision handed down after the primary authority that narrowed its application to specific facts. They will not find the string of single-judge decisions that has consistently refused to extend a proposition beyond its original context. Opposing submissions are curated documents. Counsel has chosen the cases that carry the proposition at the level of generality their argument requires. What does not appear — the narrowing authority, the distinguishing line — is not there by accident. Testing the submission means reading through its curation, not merely around it.
Generic AI does not close this gap and, on the verifiability question, sharpens it. A general-purpose model fed these submissions will return something fluent: a counterargument-shaped response, a confident summary of the legal landscape, plausible-looking citations. No audit trail. The authorities may be real, approximate, or fabricated. For a litigator preparing oral argument, fluency without traceability is a liability. Practice Note SC Gen 23, which took effect in New South Wales on 3 February 2025, and the Federal Court's April 2025 statement on responsible AI use both require the same thing: every legal citation and legislative reference verified by the practitioner. A tool that cannot show its sources cannot satisfy that obligation, however confident its output reads.
She needs a way to run the stress-test systematically — against the full current state of the relevant Australian case law, not against what she already knows — and she needs the results to be traceable, because verification is her obligation, not the tool's. This is exactly the constraint Habeas was built around: outputs grounded in a closed dataset of legitimate Australian legal sources, every result linked to its origin so that confirmation is efficient rather than aspirational.
She feeds the submissions into a Research Assistant session and runs three queries: test the authority cited on balance of convenience against the current state of the case law; identify recent decisions that have qualified or departed from the primary cases relied on; surface authorities conspicuously absent from opposing counsel's submissions. That last query is often where the real leverage lies.
The analysis comes back structured and cited, drawn from Habeas's corpus of over 300,000 Australian cases and pieces of legislation, with every proposition traceable to its source. The third authority — the one she did not know — has been distinguished twice in subsequent Federal Court decisions on facts broadly similar to those in the matter. A line of single-judge decisions over the last eight months has consistently declined to apply its balance of convenience formulation to commercial injunction applications of this kind. Opposing counsel's submissions cite none of this. The absence is now documented.
We hear this described as foundational research processes that used to take a full morning completing in minutes. That compression is what makes the 4:47pm scenario workable. She cross-checks the two most significant authorities directly, as SC Gen 23 requires, and they hold. She arrives at the hearing with a prepared response rather than an improvised one. The question she puts to the court on balance of convenience is calibrated precisely to the narrowing the other side did not cite. That is the work that requires a barrister. Habeas absorbed the manual labour upstream of it.
A Research Assistant identifies and surfaces; it does not weigh how a particular judge is likely to receive a line of argument, assess the strategic significance of what it found, or decide whether a technically weaker proposition carries practical weight on these specific facts. The analysis feeds the practitioner's thinking; it does not substitute for it. The verification obligation also stays with the practitioner, and Habeas provides traceable citations specifically so that obligation is efficient to discharge. Moving quickly from an AI-assisted research sweep to direct confirmation of each authority is a materially different position from receiving fluent but unverifiable output from a general-purpose tool.
What changes is comprehensiveness. A stress-test that previously depended on memory, conference, and available time can now be systematic, grounded in the full current state of the relevant Australian case law, and completed before the afternoon is gone. The vulnerabilities in opposing submissions are still there to be found. Habeas makes more of them findable, and makes it harder to miss the one that matters.
See it at habeas.ai.
The legal research in this article was conducted and every citation verified using Habeas, the Australian legal AI research platform.
Hero image: Vitaly Gariev on Unsplash
