A client arrives with a five-case summary their AI tool produced in thirty seconds. Four of the cases are real. The fifth doesn't exist. The client has no idea which is which, and nothing in the prose marks the gap.
The lawyer's job, suddenly, is to verify rather than originate.
To feel what that means, start with a construction dispute. The client has used AI to research their entitlement to an extension of time following a latent condition event. Their summary cites three decisions for the proposition that the principal's obligation to issue an extension was non-discretionary. Two cases support a narrowed version of that proposition. The third turns on notice provisions the client's contract doesn't contain. The AI prose doesn't flag any of this. It reads as settled. The practitioner who relies on the summary without pulling the judgments advises on facts that don't quite map to the authority, and that's before the fifth case, the one that doesn't exist, enters the picture at all.
A Lawyers Weekly op-ed framed this dynamic as a billing problem: lawyers are spending time on a new category of work, auditing client-produced AI research rather than building from scratch. How do you bill for that? The prior question is how long verification actually takes.
Because what verifying client AI research requires is this: confirm each case exists, that it says what the summary claims, that it hasn't been distinguished or overruled, and that the overall argument holds against the actual legislative text. The client's framing may be coherent but subtly wrong. A case might be real authority but cited for a proposition that represents the headnote rather than the actual reasoning. Courts regularly qualify principles in later decisions; a client AI tool working from a training corpus has no reliable way to surface that subsequent history. Legislative provisions get amended. A New South Wales decision gets treated as persuasive in a context where the Victorian statutory framework differs materially, without the qualification appearing anywhere in the prose.
These failures are more dangerous than the hallucinated case, because they look right.
Whatever research a client produces, the practitioner who relies on it accepts full professional responsibility for what it says. The Federal Court's Generative AI Practice Note, in force since April 2026, makes explicit what was always implicit: the lawyer who signs a document must personally verify that cited authorities exist and support the stated proposition. The underlying professional obligation applies everywhere. Verification is not optional overhead. The question is only how long it takes.
So the obligation is fixed. The only variable is time, and here the volume problem arrives. That client with five cases is not arriving once. They arrive every week, and each time the prose is just as fluent and just as confident. They've hit no friction points in their research. The friction is now yours.
Manual re-derivation doesn't scale against that. A practitioner working through a contested AI summary by hand, pulling up each case, reading the relevant paragraphs, checking subsequent history, cross-referencing the statute, is doing careful work. It's also work that takes a morning. There is a compounding problem: if one case in a five-case argument is wrong or mischaracterised, the argument built around it may be structurally unsound. The practitioner can't patch the bad citation and leave the surrounding reasoning intact; they may need to rebuild the analysis from scratch. A fabricated case in the controlling position of a multi-step argument is worse than one cited for a peripheral proposition. Verification, in those circumstances, slides into re-derivation regardless.
The op-ed proposed treating verification as a billable category and explaining its value to clients. That sidesteps the mechanic: if verification is slow, no billing frame resolves it. The client who feels they've already bought the time back via AI is not persuaded by an invoice line. A client who commissioned research from a law firm and received back an invoice for checking that research would find it self-evidently absurd. The experience with client-produced AI research is not entirely different from their perspective: they did the research; they've now been charged to have it verified. The conversation about the invoice is sometimes more expensive than absorbing the verification time.
Speed of verification is what changes the dynamic. When a practitioner can confirm or correct AI-sourced claims within the same client meeting, pulling up the cases, reading the key passages, identifying where the summary has drifted from what the authority holds, they occupy a different professional position. A practitioner who identifies, across the table, that one of the client's five cases has been overruled and two others are distinguishable on the notice point has demonstrated something hard to invoice for and harder to argue with.
Habeas was built for this kind of work in the Australian jurisdiction. The Search Engine covers more than 300,000 Australian cases and pieces of legislation, grounded in a closed corpus of verified primary sources, with citations traceable to the original document rather than generated from inference. When a client arrives with a proposition about the construction of a contractual term or the scope of a statutory duty, the relevant material surfaces in seconds. The citations resolve. The practitioner sees immediately whether the client's framing holds, where it overstates, what the authority says in context. The traceable citation is the operative feature: a verification workflow built on a tool that might itself hallucinate has moved the problem one layer back, not solved it. When Habeas returns a result, there is a document behind it that a practitioner can read. When the result isn't there, that's information too.
A General Counsel 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." That is the practitioner position the billing conversation ultimately requires: demonstrated authority rather than explained time.
The verification tax makes sense as a concept when verification is slow. The verification tax dissolves when verification takes minutes.
habeas.ai, book a demo.
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: Gabriel Grip on Pexels
