Law firm Sullivan & Cromwell apologized for AI errors in a court document 🤖

The story of AI reached even the place where the cost of a mistake is especially high – legal documents. The American law firm Sullivan & Cromwell was forced to apologize to the court after errors generated by artificial intelligence were found in a submitted document. For the market, this is an important case not only about law, but also about the limits of trust in AI in sensitive areas.

What happened

It concerns a court document in which incorrect references and other errors appeared. Firm partner Andrew Dietderich admitted that Sullivan & Cromwell’s internal rules regarding the use of AI in this case were not followed, and that the additional review did not work as it should have. After that, the firm submitted a corrected version of the document.

Why this matters

The problem here is not only in the fact of the error itself. The legal sphere stands on precision of wording, correctness of citation, and procedural discipline. If an AI error makes its way into a court document even in a large and experienced firm, it immediately strikes at trust in the whole process of using such tools in law.

• AI errors made it into an official court filing
• the firm’s internal review rules were not followed
• the incident showed that even large law firms are not protected from such failures

What this means for the market

For business, this is another signal that AI does not cancel human responsibility for the final result. On the contrary, the more sensitive the field, the stricter the control must be. In the case of law, an error in a text – is no longer just a technical failure, but a risk to reputation, the process, and trust in the firm itself. And that is exactly why the Sullivan & Cromwell incident reads more broadly than as a separate mess-up by one team.

Conclusion

The Sullivan & Cromwell case shows a simple thing: AI can speed up work, but it cannot replace verification discipline where law is involved. If a company has safety policies but does not follow them in practice, the risk of error quickly becomes public and costly. For the market, this is another reminder that in working with AI, the main thing is not the tool itself, but how exactly it is controlled.