A top Wall Street law firm, Sullivan & Cromwell, filed a court motion packed with errors from artificial intelligence. Opposing lawyers spotted the mistakes right away. Bogus citations. Misquoted statutes. Nonexistent cases. The firm, known for advising giants like OpenAI and representing Donald Trump, had to scramble.
On April 9, 2026, Sullivan & Cromwell submitted an emergency motion in the U.S. Bankruptcy Court in Manhattan. The case: In re Prince USA, tied to a Cambodian scam conglomerate’s collapse. Boies Schiller Flexner, the other side, flagged around 40 inaccuracies. Andrew Dietderich, co-head of the firm’s restructuring group, wrote Judge Martin Glenn on April 18. “We deeply regret that this has occurred,” he said in the letter, available at this court document. The errors stemmed from AI “hallucinations.”
Sullivan & Cromwell holds an enterprise license for OpenAI’s ChatGPT. Firm policies demand training on AI use. Those rules weren’t followed here. A secondary review missed the fakes too. The firm filed a corrected version. Now it’s probing how this slipped through.
This isn’t isolated. Back in 2023, Mata v. Avianca set the precedent. Lawyers Steven Schwartz and Peter LoDuca cited six fake cases from ChatGPT: Varghese v. China Southern Airlines, Shaboon v. Egyptair, others. Judge P. Kevin Castel hit them with a $5,000 fine, joint and several with their firm. “Acts of conscious avoidance and false and misleading statements to the Court,” he ruled, per the opinion at Berkeley Law’s archive. They doubled down at first, submitting AI-forged opinions.
Fast forward to February 2025. Morgan & Morgan, the massive personal injury outfit, faced trouble in Wadsworth v. Walmart. Attorney Rudwin Ayala used the firm’s MX2.law tool—powered by ChatGPT—to beef up motions in limine. Eight of nine citations? Phantom cases. Judge Kelly H. Rankin sanctioned Ayala $3,000, revoked his pro hac vice status. T. Michael Morgan and Taly Goody each paid $1,000. Total: $5,000. The firm withdrew the motions, paid fees, added training. Still, Rankin quoted Mata: “A fake opinion is not ‘existing law.'” Details in the order, covered by Bloomberg Law.
Judges keep piling on. In a 2025 Alabama case, Johnson v. Dunn, a firm got sanctioned; monetary penalties weren’t enough, per analyses like Jurvantis.ai. Cozen O’Connor lawyers had to write their law school deans. Courts reference Mata routinely now. Databases track over 1,300 AI hallucination incidents in filings, mostly from pros, not just pro se litigants.
But. Some firms dodge the AI label. A Liberty Mutual lawyer cited fakes; court called it “egregious human error,” fined $1,000 plus fees. No tech involved, she claimed. Above the Law mocked Sullivan & Cromwell’s plea in its April 2026 post: “Emergency ‘Please Don’t Sanction Us.'”
The New York Times broke Sullivan & Cromwell’s apology on April 21, 2026, noting the firm’s elite status in its article. The Guardian followed April 22, linking to Prince Group’s scam backdrop in its coverage. Futurism called it a “rookie mistake” for a firm repping Trump, in its April 26 piece.
AI tools promise speed. Lawyers crave efficiency amid billable pressures. ChatGPT drafts briefs fast. But verification lags. Stanford studies show large language models fabricate 75% of court ruling citations. MIT found AI 34% more confident when wrong. Courts demand Rule 11 compliance: reasonable inquiry before filing. AI skips that if unchecked.
Firms respond. Morgan & Morgan warned its 1,000+ lawyers post-sanction, per Reuters. Sullivan & Cromwell reviews policies. Bar associations issue ethics guides. Some courts require AI disclosure. Yet incidents climb. X posts tally hundreds by 2026.
Opponents weaponize it now. Spot hallucinations, tattle to the judge. New gotcha in litigation. BigLaw chases AI deals—Sullivan & Cromwell advises OpenAI—while eating its own cooking. Irony bites.
Remedies emerge. Tools verify citations across models. Firms train on tech competence. But humans sign filings. Duty stays non-delegable. AI assists. Lawyers vouch.
Sanctions deter. Fines. Bar referrals. Case ejections. Reputations dent. Elite or not, no one’s immune. The Prince USA filing? Corrected, but trust erodes. Next motion, eyes scan for fakes first.
Courts adapt slower than tech. Standing orders warn against unverified AI. Some ban it outright. Most? Expect competence. As Futurism put it: “The new courtroom gotcha? Check whether your opponent left in brainless AI hallucinations.”
Legal pros face a shift. AI boosts output. Hallucinations risk malpractice. Balance tips on verification. Firms that master it thrive. Slackers pay—literally.
WebProNews is an iEntry Publication