AI Lawyer Hallucinations: How to Verify Cases and Citations

AI lawyer hallucinations are invented case citations, false statutory quotations, and made-up authority that an AI tool presents as real. They are a known failure mode in every major AI research tool, including those designed specifically for legal work. Detecting and checking them before filing is the lawyer's responsibility, not the tool's. Using fenbs to log verification tasks and record lessons from errors helps legal teams build a reliable check into every workflow.

Courts have sanctioned lawyers who submitted AI-generated briefs with fabricated citations. The cases are public and increasing. A verification habit is not a sign of distrust in your tools—it is professional competence.

What AI hallucinations look like in legal work

A hallucinated case typically has a plausible-looking citation: a real court, a real year range, real parties. The case does not exist, or a case with those parties decided something different in a different year. The AI did not lie deliberately—it generated the most statistically likely-sounding text for the context.

Hallucinations also appear as invented statutory language ("42 U.S.C. § 9999 provides..."), fictional regulatory guidance, and real cases cited for propositions they do not actually support. The last type is especially dangerous because the case exists and looks verified, but the principle attributed to it does not appear in the decision.

Why AI generates invented citations

AI language models predict the most likely next token given prior text. A legal research prompt creates pressure toward a legal-sounding result—case name, citation, court, year, holding. When the model does not have a suitable real case in its training data, it generates a plausible-sounding one instead. There is no internal alarm that fires when this happens.

Even legal-research-specific AI tools can hallucinate, though the frequency varies. Do not assume that a tool marketed for legal research is free of the problem. Verify every primary authority it produces, regardless of the source.

How to check a case citation

For every case the AI cites: (1) look it up in a primary legal database such as Westlaw, Lexis Advance, Google Scholar, or the court's own PACER docket; (2) confirm the case exists with the exact citation given; (3) read the relevant portion of the decision yourself; (4) verify that the holding or language the AI attributed to it is actually there.

• Does the case exist? Search for the exact citation (volume, reporter, page) in a primary database.
• Do the parties match? AI sometimes applies a real citation to different parties.
• Does the holding match? Read the decision; do not rely on the AI's summary of it.
• Is the case still good law? Check for subsequent history, including reversal or overruling.

Record which cases you verified and what you found. That record protects you and helps the next person working on the same matter start from your confirmed findings rather than re-checking from scratch.

How to verify statutes and regulations

For statutory citations, check the current text of the law at its primary source. For federal statutes, the Office of the Law Revision Counsel publishes the official U.S. Code at https://uscode.house.gov/↗ (opens in a new tab). For federal regulations, check the Electronic Code of Federal Regulations at https://www.ecfr.gov/↗ (opens in a new tab). State codes and regulations have official websites; verify against those, not secondary summaries.

AI can misquote current statutory text, attribute language from a repealed version, or cite a section that was renumbered. Read the statute directly. A difference of a single word or subsection number can change the legal effect entirely.

Building verification into your workflow

Treat verification as a named step, not an afterthought. When AI assists with research, create a task or checklist item for citation verification before the work can move to drafting. Assigning the step to a named person makes it harder to skip under time pressure.

Track errors and near-misses. When an AI citation turns out to be fabricated or mischaracterized, write a brief lesson: what the AI claimed, what the primary source actually said, and what check caught it. Sharing those lessons with the team reduces repeated exposure to the same failure pattern. A record of verified sources and caught errors is also useful evidence of due diligence if a question arises.

How to do this with fenbs

Set up a verification workflow for each matter or practice area that uses AI research.

1. Create a free account and set up a project for the matter or practice area. Add a description of the research scope and any AI tools in use.
2. Write the verification requirement in Decisions and rules: all primary authority must be checked against a primary database before filing. Name the person responsible.
3. Create a task for each batch of AI-generated research. Add a checklist: each citation checked, each statute read directly, verification complete.
4. When you find a hallucination, add a lesson: what the AI cited, what the source actually says, and which check caught it. Future AI sessions on similar work read those lessons first.
5. Link to your verified primary sources in the task notes so the next person working on the matter starts with the confirmed citations rather than re-running the check.

fenbs does not connect to Westlaw, Lexis, or any legal database. It holds the governance record: verification tasks, firm rules, and lessons that help the team avoid repeating the same errors. For broader questions about professional responsibility and AI, see AI lawyer ethics.