1. Fabricated Case Citations in Filed Briefs
The most documented form of AI hallucination in legal practice is the fabricated citation. Attorneys using general-purpose AI tools to generate research have filed briefs citing cases with authentic-sounding names, real court identifiers, and plausible holdings — all invented by the AI. When opposing counsel or the court attempts to verify the citations, they cannot be found because they do not exist.
Courts in multiple jurisdictions have imposed sanctions ranging from monetary penalties to requirement of ethics training to referral to state disciplinary authorities. In several cases, the sanctions were imposed not just on the attorney who filed the brief but on the supervising partner or firm as well, on the theory that supervision obligations extend to AI-generated work product.
Prevention: Run every case citation through a verified legal database before filing. This takes thirty seconds per citation. There is no shortcut that is worth skipping this step.
2. Mischaracterized Holdings Used in Arguments
A subtler and harder-to-detect form of AI hallucination occurs when the AI correctly identifies a real case but inaccurately describes what it held. The AI might describe a case as establishing a broad constitutional rule when the court's holding was actually narrow and fact-specific. Or it might characterize a ruling as favorable to the defense when the court's actual language was more equivocal. The citation checks out; the analysis does not.
This type of hallucination causes problems when opposing counsel has actually read the case. A brief built on a mischaracterized precedent can be dismantled at oral argument by an opponent who points to the actual language of the opinion. The damage is not just to that argument — it is to the attorney's credibility on every other point in the case.
Prevention: For every case you cite as authority for a specific proposition, read the relevant portion of the actual opinion. Prefer AI tools that show grounding scores so you know which analyses to scrutinize most carefully.
3. Outdated Law Presented as Current
AI models trained on data with a cutoff date may confidently describe legal standards that have since been overruled, superseded by statute, or modified by subsequent decisions. A model trained before a significant circuit court decision or a Supreme Court ruling will have no knowledge of the change and will continue presenting the old standard as current law.
This is particularly dangerous in areas of law that have evolved rapidly — Fourth Amendment doctrine, Title IX standards, employment discrimination frameworks, and technology-related privacy law have all seen significant shifts in recent years. An AI presenting 2021 doctrine in 2025 is not hallucinating in the traditional sense, but the effect on an attorney who relies on it without checking is the same: incorrect legal analysis.
Prevention: Always check the currency of any legal standard against a verified database with current coverage, regardless of how confidently the AI presented it.
4. Fabricated Statutory Language or Regulatory Text
General-purpose AI tools can generate plausible-sounding statutory language that does not match the actual text of the statute. A model asked to summarize what a specific code section says may produce a paraphrase that introduces elements not present in the actual text, omits qualifications that are critical to the provision's application, or blends language from different sections of the code.
Unlike case citation hallucinations, statutory misquotations can be harder to catch because the correct text is often less familiar to the reader than the AI's polished paraphrase. An attorney who does not have the statute memorized may not recognize that the AI's version differs from the official text.
Prevention: Never quote statutory language from an AI summary. Always retrieve the actual text from an official source and quote directly from it.
5. Fabricated Procedural History or Expert Authority
Beyond case law and statutes, AI tools can hallucinate procedural history — describing lower court rulings, interlocutory appeals, or prior proceedings in a case that did not happen — and can fabricate expert citations, describing academic articles, treatises, or law review commentary that does not exist. Both types of hallucination have appeared in legal documents and have required correction or sanctions when discovered.
The common thread across all five of these failure modes is the same: AI confidence does not equal AI accuracy. The appropriate response is not to abandon AI tools — they provide real research efficiency gains — but to use them with verification workflows matched to their specific failure modes. Tools that show you grounding scores, cite their sources, and surface the underlying documents make that verification faster and more systematic.
Prevention: Treat all AI-generated factual claims — not just case citations — as hypotheses requiring verification. Use tools that show their work.