Cheaper Isn’t Always Better: US Judge Says Clients Should Think Before Requiring Lawyers to Use AI for Briefs

A Manhattan federal judge declined to sanction lawyers over AI-linked citation errors while warning that cost savings must be weighed against accuracy and lawyer training.

What Happened

US District Judge Arun Subramanian of the Southern District of New York has warned lawyers and their clients to think carefully about how artificial intelligence is incorporated into legal work after a court filing contained citation errors linked to AI.

According to Reuters, Subramanian declined to impose sanctions on the lawyers involved in the copyright case. Instead, he used the episode as a warning for the profession, stressing that AI-generated legal material must be independently checked before it reaches a court.

The judge also focused on a question that extends beyond the lawyers who filed the document: what happens when clients want law firms to use AI primarily to reduce legal costs?

According to the reported decision, Subramanian observed that clients may seek AI adoption for efficiency and cost savings, but that mistakes can produce consequences that outweigh those savings. He suggested that clients may therefore want to consider carefully whether they should require lawyers to use AI for tasks such as preparing legal briefs.

The remarks did not amount to a prohibition on AI use. Subramanian reportedly acknowledged that AI may become an increasingly unavoidable part of legal practice. His concern was how firms and clients manage the risks that accompany that transition.

Background & Context

The dispute arose in a copyright infringement case in New York in which Foundation Media was the defendant.

On September 22, 2026, Subramanian questioned Foundation Media’s lawyers about the use of AI in a court filing. Among the issues was whether AI had generated certain quotations and a citation that had not been checked before the filing was submitted.

Cynthia Arato of Shapiro Arato Bach, which represents Foundation Media, apologised for the mistakes. According to a declaration she submitted, another lawyer at the firm had prepared and finalised the filing contrary to the firm’s policies and training requirements.

Arato’s declaration attributed the problem to a failure to verify AI-generated citations. She characterised it as an isolated error and said it was not the result of bad faith or an intention to mislead the court. The firm said it was implementing additional safeguards against similar misuse.

A later filing disclosed that the firm had used AI tools built by Harvey in preparing the filing. The reports do not establish that a particular Harvey tool produced the errors.

The judge ultimately decided against imposing sanctions.

Key Details

• Court/Forum: US District Court for the Southern District of New York, Manhattan, before Judge Arun Subramanian.
• Date: Reuters reported the decision on October 2, 2026. The judge had questioned counsel about AI use on September 22. The date of the reported decision itself is not specified here.
• Case: A copyright infringement case involving Foundation Media as defendant. The case title and docket number are not included because they were not part of the confirmed reporting used for this article.
• Counsel: Cynthia Arato of Shapiro Arato Bach represented Foundation Media.
• AI Tool Disclosed: A later filing said the firm had used tools built by Harvey while preparing the filing.
• Outcome: No sanctions were imposed.

Why It Matters

The reported remarks put three separate issues on the table: verification, client choice and lawyer training.

First, lawyers remain responsible for the material they place before a court. The fact that AI may have assisted with research or drafting does not remove the need to check citations, quotations and other substantive material before filing. The episode is therefore less about whether AI can be used than about what level of human verification accompanies its use.

Second, the judge’s comments introduce a practical question for clients. A client may reasonably want a law firm to adopt technology that promises faster work or lower costs. But if the technology introduces errors that require correction, create procedural complications or undermine a filing, the economic calculation can change.

That does not mean clients should avoid AI or that AI-assisted legal work is inherently improper. Rather, the reported observation asks clients to consider whether they should require AI use for particular tasks, especially where the potential consequences of an error are significant.

For law students and junior lawyers, there is a third issue: training.

Subramanian reportedly warned that heavy reliance on AI could reduce opportunities for junior lawyers to develop the skills that traditionally come from preparing legal research and drafts. He suggested, as one possible approach, requiring initial drafts of briefs to be prepared without AI assistance so that junior lawyers continue to learn research, analysis and drafting before technology is introduced into the workflow.

That suggestion is not a judicial rule or binding direction. It is part of the judge’s reported discussion about how the profession might preserve training while adopting increasingly capable technology.

For law firms, the resulting question is therefore broader than whether an AI system can complete a task. It is how the technology fits into supervision, verification, client expectations and professional development.

For clients, the decision also highlights a less obvious part of the AI-cost debate. The cheapest workflow at the drafting stage may not necessarily be the lowest-risk workflow overall if errors create additional legal or professional consequences.

Reactions

Cynthia Arato apologised for the filing errors and, according to her declaration, described the failure to verify the AI-generated citations as an isolated mistake made without bad faith or an intention to mislead the court.

She also said the firm was implementing additional safeguards against AI misuse.

According to the reports, neither Harvey nor Foundation Media responded to requests for comment.

The firm’s explanation remains its own account of what happened. The reported decision to impose no sanctions is separate from that explanation and does not establish that every detail of the firm’s account was independently verified by the court.

Closing

The episode does not establish a rule against AI-assisted legal work. Nor does the judge’s warning amount to a direction that firms or clients should stop using AI.

Instead, the reported remarks highlight a developing tension in legal practice: clients may want technology to make legal services faster and cheaper, while lawyers and firms remain responsible for the accuracy of what ultimately reaches the court.

For junior lawyers, the same transition raises a different question: if AI performs more of the basic drafting and research work, how will new lawyers acquire the skills that those tasks traditionally teach?

The judge’s answer, at least as reported here, is not to reject AI but to think more carefully about where it enters the legal workflow, who verifies its output and how the profession continues to train its next generation.