Hallucination
A hallucination is output from an AI system that is fluent and plausible but factually false — in legal contexts, most consequentially an invented case, citation, or statutory provision.
In plain English
Hallucination is a consequence of how language models work: they generate the most probable continuation of text, not a lookup from a verified database. Where the format of an answer is highly patterned — as legal citations are — the model can produce something perfectly shaped and entirely fictitious. Retrieval grounding reduces this substantially but does not eliminate the need for verification.
Why it matters
This is the central risk that determines whether legal AI is usable professionally. It is also why architecture matters more than model choice: the mitigation is source grounding plus citation verification plus a human review step, not simply a better model.
Example
Asked for authority on a point, an ungrounded model returns a case name, court, year and citation in perfect format. None of it exists. The output is indistinguishable from a real citation without checking it against a real corpus.
How LexVio handles it
LexVio grounds contract analysis and research in indexed Indian statutes and case law, and returns linked source judgments rather than unsourced assertions.
LexVio — Legal AI PlatformCommon questions
How do you prevent AI hallucination in legal research?
Ground answers in retrieval from an authoritative corpus, cite the specific source and paragraph, verify that each citation exists and remains good law, and keep a human review step before the output is relied on.
Does retrieval eliminate hallucination?
It reduces it substantially but does not eliminate it. A model can still mischaracterise a retrieved source, so citation verification and human review remain necessary.