AI-Generated ‘Phantom’ Cases Haunt U.S. Courts, Lawyers Urged to Verify Every Citation

AI Hallucinations Are Filling Legal Filings with Nonexistent Cases

American courts are grappling with an alarming side effect of the legal profession’s rush to embrace generative artificial intelligence: a growing number of motions, briefs, and pleadings that cite court rulings that simply do not exist. Lawyers, relying on AI tools to accelerate research and drafting, are unintentionally submitting fabricated case names, quotes, and citations—AI “hallucinations” so convincing that they go unnoticed until judges or opposing counsel try to track them down.

The phenomenon has triggered stern warnings from federal and state judges, disciplinary bodies, and legal ethics experts, all of whom emphasize that the duty to verify every authority before filing remains squarely with the attorney, no matter how the text was produced.

The Double-Edged Promise of Legal AI

Crawford Appleby, a partner at Wisner Baum, told Federal News Network that artificial intelligence can dramatically speed up the early stages of legal research and drafting, helping firms handle large-scale litigation and document review. Yet he stressed that the technology’s ease can create dangerous blind spots. “The tools are powerful,” Appleby said, “but they require rigorous human oversight. Any lawyer who files a document without personally checking every citation is rolling the dice with their client’s case and their own license.”

The core risk is the tendency of large language models to generate plausible-sounding but entirely fictitious authorities. An AI might return a perfectly formatted case caption, complete with a realistic docket number and a coherent-sounding legal holding—none of which correspond to an actual decision. When a lawyer trusts that output without verification, the fabricated material enters the public record, wasting judicial resources and undermining the credibility of the filing.

Courts and Regulators Push Back

Courts are not treating these missteps as harmless errors. Multiple judges have issued show-cause orders demanding explanations from attorneys who submitted briefs peppered with phantom cases. Bar regulators and judicial committees, guided by evolving ethics opinions, have made clear that inaccurate AI-generated citations are a serious professional responsibility failure.

The American Bar Association’s ethics guidance urges lawyers to maintain active oversight over any technology used in practice, and several federal courts have updated their local rules or standing orders to require lawyers to disclose the use of AI and to certify that all citations have been independently verified. A failure to do so can result in sanctions ranging from monetary fines to referral to disciplinary authorities.

How Law Firms Are Responding

In response to the rising number of AI-citation failures, law firms—both large and boutique—are rapidly adopting internal safeguards. Common measures now include:

• Mandatory citation-checking protocols that require every case reference to be verified against the official reporter or court docket before filing.
• Limits on unsupervised AI use, confining generative tools to first-draft drafting or research tasks that an attorney must always re-check.
• Training programs that teach legal professionals to recognize AI hallucinations and understand the ethical duties that cannot be delegated to software.
• Internal review layers, where senior attorneys or dedicated legal-research librarians audit filings flagged as AI-assisted.

Appleby noted that Wisner Baum has instituted similar checks, viewing them as non-negotiable. “The time saved by AI means nothing if a filing gets tossed out or a lawyer faces an ethics complaint,” he said.

A Broader Debate on Safe AI in Law

The fake-citation problem sits within a wider conversation about how generative AI should be integrated into legal practice. While many practitioners embrace the technology for tasks like contract analysis and e-discovery, the evident risk of fabricated output has hardened the consensus that human oversight must remain mandatory at every step that touches the court.

Legal technology vendors are responding with features designed to ground outputs in verifiable sources, and court administrators are studying whether to build automated citation-checking tools into the electronic filing system itself. For now, however, the burden—and the liability—rests completely on the attorney who presses “file.”