Sullivan and Cromwell, one of the most powerful law firms on Wall Street, apologized to a federal bankruptcy judge this month after submitting a court filing riddled with inaccurate citations and errors generated by artificial intelligence. The same week, a California lawyer named Steve Brigandi was fined nearly $100,000 after submitting dozens of briefs containing fabricated case law in a $12.6 million inheritance dispute over an Oregon winery. A separate New Jersey lawyer was fined $5,000 for his second AI-related offense. And California's State Bar charged two more lawyers and approved discipline for a third over AI-generated nonexistent citations.

The AI hallucination problem in courts has reached a new threshold. It is no longer solo practitioners or careless associates making these mistakes. It is now Sullivan and Cromwell, a firm that bills over $2,000 per hour and has represented Goldman Sachs, JPMorgan, and the U.S. government.

The Sullivan and Cromwell Filing

As reported by Reuters and The New York Times, a senior Sullivan and Cromwell lawyer discovered on April 17 that an emergency motion filed with the court on April 9 contained multiple inaccurate citations and errors generated by AI. The firm submitted a letter to the judge identifying and correcting each error in a detailed schedule. The apology was direct: the filing "includes inaccurate citations and other errors." Sullivan and Cromwell did not disclose which AI tool produced the errors or explain how they bypassed the firm's review process.

This is the highest-profile firm to be publicly caught. Sullivan and Cromwell is not some mid-tier outfit fumbling with new technology. It is a 147-year-old institution that advises on the largest corporate transactions in the world. If their review process cannot catch AI hallucinations, the question becomes: whose can?

The Oregon Winery Case

The Brigandi case may be more alarming in its details. As reported by The New York Times and Jezebel, the dispute centered on Valley View Winery in Oregon, a family business established in 1972. After the family patriarch died, two brothers ran the winery for decades. Their mother later transferred ownership to them in estate documents. But before her death in 2023, she moved in with another sibling, Joanne Couvrette, and filed a new estate plan redirecting the winery to Couvrette and a fourth sibling. The resulting lawsuit sought $12.6 million in damages.

Couvrette's lawyer, Brigandi, was hired because her daughter was dating his son. He agreed to represent her for free. Almost immediately, filings containing AI-hallucinated legal citations began appearing: two in a January 2025 filing, seven in April, and 16 more in May. Opposing lawyers flagged the fake citations repeatedly. The filings kept coming.

A judge found "persuasive" evidence suggesting the briefs were actually written by Couvrette herself using AI, not by Brigandi. The filings contained irrelevant citations to free speech cases, which happened to relate to a separate lawsuit Couvrette was pursuing after being fired for calling pro-Palestinian protesters "terrorist sympathizers." The AI tool appeared to be pulling research from her other case and inserting it into the winery dispute.

The judge permanently dismissed Couvrette's case and fined Brigandi nearly $100,000, what appears to be the largest monetary sanction yet imposed on a lawyer for AI-related court submissions.

The Numbers Are Getting Worse

A database maintained by legal researcher Damien Charlotin tracking court cases derailed by AI hallucinations has grown rapidly. U.S. judges have sanctioned lawyers in dozens of cases after attorneys used AI for research and drafting without verifying the results. California's State Bar is now actively pursuing disciplinary action. A New Jersey lawyer was fined $5,000 this month for his second AI offense, with the judge warning she "will not hesitate" to refer him to the Disciplinary Board for a third.

The pattern is consistent across every case: lawyers use AI chatbots for legal research, the AI generates plausible-sounding but entirely fabricated case citations, and the lawyer submits them without checking whether the cases actually exist. The consequences range from embarrassment to case dismissal to six-figure fines.

What Comes Next

Courts have responded with a patchwork of disclosure requirements. Some federal judges now require lawyers to certify whether AI was used in preparing filings. But there is no uniform standard. The American Bar Association has issued guidance but no binding rule. State bars are moving at different speeds, with California leading on enforcement.

The Sullivan and Cromwell incident changes the calculus. This is no longer a story about incompetent solo practitioners or negligent associates. When a firm of this caliber submits AI-generated errors to a federal judge, the profession has a systemic problem. The technology that was supposed to make lawyers more efficient is now generating liability faster than it generates billable hours.