Sullivan & Cromwell AI Error: Inside the Court Apology Saga

Sullivan & Cromwell AI error exposed 42 flawed citations in a court filing. Explore the apology, AI risks, and legal lessons now!

In April 2026, Sullivan and Cromwell submitted. A bankruptcy court filing consisting of 42 Powered by AI errors, Fabricated case citations, Incorrect laws, and references to cases that do not exist.

The organization apologized in writing. The judge, Admitted its own AI policies Not followed, and corrected. The filing. This is it. The full story, in plain language.

I laughed out loud the first time I read the headline. Not because it’s funny , it isn’t, really , but because of the irony. This is a firm where partners bill north of $2,000 an hour. A firm that’s represented U.S. presidents.

A firm which sits on top. The legal food chain. And here they were, sending. A letter To a federal judge As basically said: our AI Made things up, and no one caught it until we filed it away.

I’ve followed Legal Updates and legal tech stories for a while now, mostly because I once worked alongside a litigation support team. I remember the exact moment “AI hallucination” stopped being conference-room jargon and became a real courtroom liability. This case is a big reason why.

Let’s walk through what happened, what the Sullivan and Cromwell letter actually said, and what it teaches anyone working near the legal profession right now.

What Happened: The Sullivan & Cromwell AI Error, Explained

Here’s the short version. The motion contained real errors, not typos or formatting slips. Reporters and court records later put the total at 42 separate inaccuracies packed into that single filing , this is the Sullivan & Cromwell ai error everyone’s now talking about.

Sullivan& Cromwell archived the motion on April 9. Go case is In re Prince Global Holdings Limited, Case no 26- 10769( MG), I heard the U. S. Bankruptcy Court to the Southern District of New York First Chief Judge Martin Glenn. Represented the company. The joint provisional liquidators liquidation Prince Global Holdings, is associated with a company Chen Zhi, Founder of the Cambodian Congregation U. S. The prosecutor was indicted. Wire fraud and money Accusations of money laundering, and by whom U. S.

And UK The governments all around approved the same period in connection With forced labor and fraud operations with cryptocurrency.

Sullivan & Cromwell’s own team didn’t catch the mistakes. The opposing side did.

Quick Facts Table

Detail	Information
Firm	Sullivan & Cromwell
Filing date	April 9, 2026
Apology letter date	April 18, 2026
Court	U.S. Bankruptcy Court, S.D.N.Y.
Judge	Chief Judge Martin Glenn
Case	In re Prince Global Holdings Limited, No. 26-10769 (MG)
Errors found	42 inaccuracies (fabricated citations, misquotes, fake authorities)
Discovered by	Boies Schiller Flexner (opposing counsel)
Author of apology	Andrew Dietderich, co-head of global restructuring
Court-imposed sanction	None reported , firm self-disclosed and corrected the filing

Boies Schiller Flexner Caught the Mistake

Sometimes your biggest wake-up call comes from the person across the table, not your own team. That’s exactly what happened here. They noticed the Bankruptcy Code was misquoted. They found legal authorities that were misdescribed. The firm owned the mistake fast.

The Sullivan and Cromwell AI Apology, In Their Own Words

Nine days after the flawed motion landed, Andrew Dietderich , co-head of Sullivan & Cromwell’s global restructuring group , sent a letter to Judge Glenn on April 18. This document is what most people mean by the Sullivan and Cromwell letter, and its directness stands out for a firm of this caliber.

Dietderich didn’t dance around the issue. He confirmed the inaccuracies included what he called AI “hallucinations” , moments where an AI tool fabricates case citations, misquotes legal authorities, or invents sources that don’t exist anywhere. The firm” deeply regrets that this has happened,” he wrote. He is also called Boies Schiller Flexner. Thank them personally for catching the issue and apologize directly. That small detail stuck with me: picking up the phone instead of hiding behind formal language says something about how seriously the corporation treated the underlying mistake. It was really embarrassing.

Here’s what makes this story bigger than “law firm makes AI mistakes,” though. Sullivan & Cromwell already had policies built specifically to prevent this exact scenario. Per Dietderich’s letter, lawyer access to generative AI tools requires completing mandatory training that “repeatedly emphasizes the risk of AI hallucinations” , training the firm says it tracks and verifies before granting access. This wasn’t a firm that hadn’t considered the risk. This was a firm with guardrails already in place, watching someone step right around them. The firm admitted plainly: its AI policies “were not followed in connection with the preparation of the motion,” and its internal review process failed to catch the errors before filing.

What Sullivan & Cromwell Did After the Apology

The firm backed its words with action.

Reviewed every filing submitted in the Prince Global Holdings matter to check for additional AI-related errors (it confirmed there weren’t any)
Submitted a corrected, red-lined version of the original motion , some sections required rewriting entire sentences
Launched an ongoing evaluation of whether its internal training and review protocols need strengthening

The letter left two things undisclosed: which specific AI tool produced the errors, and which lawyers were responsible. It also didn’t mention any internal disciplinary action. One online observer following the letter noted an interesting wrinkle worth considering: many modern AI research tools now generate citations with clickable, verifiable links, which would make an unverified hallucination even easier to catch before filing. If that verification step really was just a click away, this looks less like a pure policy failure and more like a skipped final check , which somehow makes it both more relatable and more frustrating.

This Isn’t a One-Off , It’s a Pattern

What struck me most while researching this wasn’t that it happened to Sullivan & Cromwell specifically. It’s that this keeps happening, everywhere, across the legal profession. Paris-based law lecturer Damian Charlotin maintains a database tracking confirmed AI hallucination cases in court filings worldwide. By the time this story broke in April 2026, that tracker had already logged more than 330 confirmed cases globally , and it has grown considerably since, with more recent counts running into the thousands.

Think about that for a second. AI-generated legal hallucinations don’t just hit solo practitioners cutting corners. They’re showing up at elite, white-shoe firms with mandatory AI training and dedicated review processes. Sullivan& Cromwell employs Approx 900 lawyers, I run the most profitable per partner economy. The world, and construction formal safeguards Absolutely to cease it. If it can be there, it can be almost anywhere. Judges across multiple jurisdictions already started finding attorneys Too personal these errors, And this trend shows no signs of abating.

What Legal Teams Can Learn From This

Beyond “use AI carefully” , thanks, very helpful , what’s the real, practical takeaway? A few lessons stand out:

Written policy doesn’t guarantee real-world compliance. Sullivan & Cromwell built training modules, tracked completion, and embedded explicit hallucination warnings into onboarding. None of it mattered once someone skipped the actual review step. A policy only works as well as the habit of following it under deadline pressure , and emergency motions almost always come with deadline pressure.

Treat citation-checking as a hard gate, not a soft suggestion. When an AI tool generates a citation, a human needs to physically pull up that case before it goes anywhere near a filing. Don’t skim it , open it, read it, confirm the quoted language actually appears. This sounds tedious. It is tedious. It’s also non-negotiable now.

Respond quickly with transparency. Sullivan& Cromwell Avoided being a drunken blowhard reputational disaster Mostly because he owned it. The mistake Apologised quickly, directly( incl that personal phone call), And self- reported a full review Instead of waiting to be recaptured.

Treat this as a training problem, not just a technology problem. AI tools aren’t disappearing, and honestly, they shouldn’t , used correctly, they save enormous time on legal research. But “correctly” carries real weight in that sentence. Firms need to treat AI verification the same way they treat conflict checks or privilege review: as a mandatory, non-skippable procedural step.

Frequently Asked Questions

What AI tool does Sullivan& Cromwell Use? The firm’s letter To the court Which was not disclosed. AI tools created the inaccurate citations.

How many errors did I show up in the filing? 42 defects including fabrication were found in case citations, Incorrect laws, and references to cases that do not exist.

What does the court fine or sanction Sullivan& Cromwell? No court sanction has been publicly reported. This instance. The organization itself revealed. The errors, Corrected the filing, And said it considers its internal protocols Warrant to strengthen.

Which case triggered this filing? In re Prince Global Holdings Limited , a Chapter 15 bankruptcy proceeding connected to the fraud and forced-labor case against Chen Zhi, heard in the U.S. Bankruptcy Court for the Southern District of New York.

Is this a common problem in the legal industry? Increasingly, yes. Tracking databases had documented more than 330 confirmed instances of AI-generated hallucinations in court filings worldwide by the time this story broke, with the count climbing well past that since , across firms of every size.

Key Takings

This story feels remarkably human underneath all the legal jargon. A team under pressure.
A shortcut that felt harmless in the moment. A mistake that ballooned into a very public, very avoidable embarrassment for one of the most respected names in law.
It’s a solid reminder , for lawyers, for anyone using AI tools professionally, honestly for all of us , that these systems genuinely help, but they still need a human checking their work before that work carries your name out the door.