trade crypt

AI hallucinations in legal filings: 28 erroneous citations found

HomeMarketsAI hallucinations in legal filings: 28 erroneous citations found

-

Sullivan & Cromwell has acknowledged significant errors in a legal filing related to a U.S. bankruptcy court case due to AI hallucinations in legal filings. The admission was made concerning a motion filed on April 9 in the U.S. Bankruptcy Court for the Southern District of New York. The firm noted that the errors included fabricated citations and incorrect characterizations of legal authorities. These issues were corrected following the withdrawal of the original motion and submission of a revised version.

Court-appointed liquidators from the British Virgin Islands have sought recognition under Chapter 15 to act on behalf of creditors and alleged victims linked to Prince Group and Chen Zhi. The filing identifies those liquidators as the foreign representatives in proceedings connected to Prince Group and Chen Zhi. The matter is presented within bankruptcy proceedings.

Prosecutors allege Chen Zhi directed scam operations that targeted victims worldwide and that billions of dollars in cryptocurrency are linked to the activity. Chen was detained in Cambodia and repatriated to China. U.S. authorities have linked Prince Group to large-scale fraud operations in Southeast Asia. Prince Group has been sanctioned by the UK and U.S. governments.

The filings reflect cross-border elements, including foreign liquidators and allegations involving activity in multiple countries. The allegations and sanctions are part of the record in the ongoing bankruptcy-related proceedings now.

Sullivan & Cromwell admitted to significant errors in a recent legal filing due to artificial intelligence “hallucinations.” These errors included at least 28 erroneous citations and mischaracterizations of authorities in a court document. Some quotations attributed to the court were found to be non-existent, while certain citations referred to incorrect legal circuits. This was formally acknowledged in a letter to Judge Martin Glenn in the U.S. Bankruptcy Court for the Southern District of New York.

The firm has implemented strict internal policies concerning the use of AI tools. Before access to generative AI tools is granted, Sullivan & Cromwell requires lawyers to complete two training modules, which emphasize the risks associated with AI hallucinations. These include the fabrication of case citations, misinterpretation of authorities, and inaccurate quotations. Andrew Dietderich, head of the firm’s restructuring practice, stated, “We deeply regret that this has occurred,” highlighting the critical nature of these issues and the firm’s commitment to addressing them.

The original motion was withdrawn and a revised version was filed after objections from opposing counsel at Boies Schiller Flexner. Those counsel identified at least 28 erroneous citations in the withdrawn filing and said some language attributed to the U.S. Bankruptcy Code could not be found. They also noted that the corrected submission misstated case law in multiple places and that several authorities were mischaracterized or misidentified, with some citations referring to a different circuit. The timing of the corrected submission prompted a dispute, as opposing counsel argued the revision was prejudicial because it came after they had lodged objections and after a hearing status conference had been requested.

Sullivan & Cromwell acknowledged that a recent bankruptcy filing contained AI-generated hallucinations that produced fabricated authorities and mischaracterized legal authorities, and the firm withdrew the original motion and filed a revised submission. The firm said lawyers must complete two required training modules before using generative AI, and the training emphasizes risks including citation fabrication, misinterpretation of authorities, and inaccurate quotations.

This website and its articles do not provide any investment advisory services within the meaning of applicable regulations. The information published may be incomplete, outdated, or contain errors. The author makes no representation or warranty regarding the accuracy, completeness, or timeliness of the information presented. Use of this information is entirely at the reader’s own risk. Under no circumstances shall the author be held liable for financial decisions made on the basis of the content published on this website.
Crypto Fan
Crypto Fanhttps://calipsu.com
Calipsu.com is dedicated to providing clear, reliable, and accessible information about cryptocurrencies, blockchain technology, and decentralized finance (DeFi). Its mission is to help readers better understand a rapidly evolving ecosystem that is often complex, technical, and misunderstood. The platform covers a wide range of topics, from major blockchain networks and crypto assets to DeFi protocols, Web3 applications, and emerging trends. The website also publishes practical guides and tutorials that explain how decentralized tools function, such as wallets, staking mechanisms, lending protocols, and liquidity pools. These guides aim to describe processes and risks clearly, helping readers understand the mechanics behind DeFi rather than encouraging participation.

LATEST POSTS

KuCoin EU AML hires to appease Austrian regulator (FMA): Developments

KuCoin EU AML hires to appease Austrian regulator (FMA): KuCoin EU expands its AML team appointing Carmen Kleinhans and deputies to meet demands.

Canada crypto ATM ban advances in Spring Economic Update

Canada crypto ATM ban advances in the Spring Economic Update as regulators target crypto ATMs to curb fraud and money laundering.

AI-generated websites Show 107% Higher Positive Sentiment

By mid-2025, 35% of newly published websites are AI-generated websites or AI-assisted, signaling a rapid AI-driven shift in online content.

What CFTC AI review of crypto registration applications means

CFTC AI review of crypto registration applications is accelerating reviews as the agency trims staff, boosting feedback speed and market surveillance.

Follow us

116FansLike
745FollowersFollow
148FollowersFollow
trade crypt