Version 2

AI hallucination in court filings has grown from scattered incidents into a documented systemic crisis, with a public database now tracking 1,453 cases worldwide — up from fewer than 120 just two years ago [1][2][3]. In April 2026, Sullivan & Cromwell, one of Wall Street's most elite law firms, acknowledged AI hallucinations caused errors in a high-profile court filing [8][9], demonstrating that even top-tier firms are not immune. Two active federal cases define the legal stakes: lawyers who used MarcTrent.AI face sanctions for submitting fabricated case citations [10], and OpenAI is seeking to dismiss a lawsuit claiming ChatGPT engaged in unlicensed legal practice after a former Nippon Life employee used it to flood courts with filings [15][14]. Bar associations have issued guidance on AI use, but practitioners widely find it too vague to be actionable [23][19][20].

The combination of exponential case growth, six-figure sanctions [2], and AI hallucination incidents at elite firms signals this is not a niche compliance problem but a systemic failure mode embedded across the legal profession. The MarcTrent.AI sanctions case and the OpenAI unauthorized-practice lawsuit will set precedent shaping every law firm and legal AI vendor — and the gap between vague bar guidance and enforceable rules currently leaves attorneys exposed.

AI-generated false citations in court filings have grown from a novelty into a quantifiable crisis. A public database maintained by researcher Damien Charlotin now tracks 1,453 such cases worldwide [1], having grown from fewer than 120 to over 660 in under two years [2][3]. Six-figure sanctions have already been imposed in some instances [2], and courts from the Sixth Circuit [4] to Massachusetts [5] to Pennsylvania [6] to Oregon [7] have either issued sanctions or surveyed the growing landscape of cases. In April 2026, the problem reached a new level of prominence: Sullivan & Cromwell, one of the most prestigious law firms on Wall Street, acknowledged that AI hallucinations had caused errors in a high-profile court filing [8][9], demonstrating that even elite, well-resourced practices are not immune to the failure mode.

Two concurrent federal cases are forcing courts to confront AI's role in legal practice from opposite directions. The first centers on MarcTrent.AI, a firm that markets its AI platform as boosting legal success rates by 35 percent through predictive modeling and AI-driven case discovery [10]. Attorneys who relied on its research tools submitted what appear to be fabricated case citations in a federal appeal arising from a defamation dispute in which Nikko D'Ambrosio sued more than two dozen women and Meta after a Chicago Facebook group called 'Are We Dating the Same Guy' published a critical post about him [10]. The underlying suit was dismissed with prejudice at the district court, but the citation problem surfaced on appeal, leaving the attorneys facing potential sanctions. The case exemplifies a recurring pattern: AI systems generating plausible-sounding but wholly invented legal precedents, with practitioners failing to verify them before filing [4][11][12][13].

The second active case raises a structurally distinct question about where AI liability itself sits. OpenAI has moved to dismiss a lawsuit — filed in Illinois in March 2026 [14] — that accuses ChatGPT of engaging in unlicensed legal practice after a former Nippon Life employee used the tool to generate a large volume of court filings following a disability-benefits settlement [15]. OpenAI's motion frames ChatGPT as software rather than a legal services provider [15][16], a position consistent with the company's broader stance that its models are tools rather than autonomous agents. The case has drawn academic attention as a test of where the AI-as-tool framing breaks down under unauthorized-practice-of-law doctrine [17][18].

The legal profession's institutional response has been uneven. The American Bar Association has published practical checklists for responsible AI use [19], and state bars in California [20] and Texas [21] have issued formal guidance. Industry observers like Thomson Reuters frame the hallucination problem as one solvable by better legal practice rather than structural reform [22]. But practitioners report that the existing guidance is so vague it amounts to telling attorneys to figure it out on their own [23], and no uniform national standard currently governs whether attorneys must disclose AI use in their filings. The result is a profession in which AI is deeply embedded in legal workflows, liability for AI failures falls entirely on the attorney of record regardless of vendor marketing claims, and the rules governing acceptable use remain fragmented and largely unenforceable.