Legal arguments that chain across sources — statute to regulation to the case construing it to the exception — give the model more joints to fabricate. The endpoints can be real while the connective tissue between them is invented.
Probability map. Many people are under the mistaken impression that an LLM is some sort of hypercomplex thinking machine. Really it’s just an incredibly huge probability map of words, patterns, and phrases followed by other words, patterns, and phrases. Unless it’s attached to a database of authoritative sources, nothing connects what it generates back to reality.
Completion pressure. Every model runs under a standing directive that strongly favors providing an answer rather than tell you it can’t. Hand it a gap where a real authority should be, and it fills the gap instead of flagging it.
Latent contradiction. Your LLM prompt can look fine on its face, but against the backdrop of law, non-obvious contradictions can be hiding in wait. That conflict is invisible to you, and sometimes even to the model itself. Something has to give, and completion pressure decides what: it honors the request and manufactures the rest.
Many legal conclusions require a chain: a statute delegates to an agency, a regulation implements it, a case interprets the regulation, and an exception carves out the client. Each link is a separate retrieval, and each is a separate opportunity to substitute a plausible invention for a real source.
Even when the model has the endpoints right — the correct statute, the correct general rule — it may fabricate the reasoning that connects them: a case that supposedly bridges two provisions, a subsection that supposedly contains the exception, a holding that supposedly ties the regulation to the client’s facts. The deeper the analysis, the more joints there are, and the more the errors compound.
The danger is that a chain with real endpoints looks sound. A reader spot-checking the first and last citation finds them genuine and trusts the middle — which is exactly where the invented link tends to sit.
The same cause, a few ways it turns up in ordinary practice. Each is routine work you’d never flag as risky — which is exactly how the cause slips in unnoticed.
Anywhere the argument has to travel — a statute to the case applying it to your facts, a general rule to its niche application, the authority that ties two provisions together.
The model supplies the bridge case — real-looking, correctly formatted — that doesn’t exist, sitting between two authorities that do.
Regulatory work full of nested carve-outs — a CFR part with an exception three subsections deep, a tax reg’s special rule, an agency guidance’s exemption.
It cites a subsection for the exception; the regulation is real, the subsection isn’t.
The long analogical reach — reading a decision a step past what it said, stretching a holding to cover facts it never faced, the inference the chain needs but the opinion doesn’t make.
The “holding” that completes the chain is a paraphrase the opinion never actually states.
Adding a citation directive (“only cite real cases”) mathematically cannot ever fix the problem. Humans read a citation directive and think the LLM must comply. But every LLM has completion pressure built into its architecture, a mathematical property of the model itself, that will readily override such a directive.
Verbatim checks every link independently — each case, each statute, each regulation the chain relies on, resolved against its source and matched to the quoted language. A real endpoint no longer vouches for an invented middle: the bridge case that doesn’t exist and the subsection that isn’t there surface as their own findings, however solid the citations around them look.
Every anchor in the argument is verified back to a source you can open and read, so a chain is only as trusted as its weakest verified link — not as its most reassuring one.
Verbatim reads a finished brief and reports, for every authority it cites, whether the cite is real and whether the quoted language actually appears at the pin cite — so a fabrication surfaces on your screen, not in a show-cause order. Bring a brief and we’ll walk you through the report.