AI training on proprietary legal documents

AI

Law Firm Discovers Its Paralegals Trained On Opposing Briefs

A routine audit of the firm's AI system revealed that training data collection included documents from closed litigation files.

By Nextish DeskAI
Closeup image of a law book titled 'The Law' on a wooden desk with scales of justice.
Photo by Mikhail Nilov on Pexels

A mid-size litigation firm in Chicago discovered Tuesday that its in-house AI system had been trained on thousands of opposing counsel briefs, memoranda, and settlement proposals dating back seven years, according to internal emails reviewed by Nextish News. The system, deployed last year to accelerate document review for junior associates, had ingested the firm's entire archived case file library without human review of what constituted training material. The discovery came during a compliance audit prompted by a partner's routine question about how the system knew to flag a particular litigation strategy before the firm had developed it.

Managing partner Richard Castellano, fifty-two, said in a statement that the firm had believed it was training the system on its own work product only. "We thought we were building institutional knowledge," he said.

We thought we were building institutional knowledge.

The audit revealed that the firm's IT department had configured the training pipeline to pull from every document stored on the litigation servers, including materials from cases settled or closed for more than five years. This included complete opposing counsel strategies, expert witness reports prepared by competing firms, and in one instance, the internal billing records and client communications of a former adversary whose files had been archived after settlement. The system had learned to recognize patterns in how particular opposing counsel approached motion practice, what expert witnesses they favored, and at what settlement ranges they typically folded.

An associate who helped identify the issue, speaking on condition of anonymity, said the firm's leadership seemed divided on whether to disclose the problem. "Some people said we should just not mention it to anyone, because technically we had legal possession of the documents," the associate said.

The firm has since retained outside counsel to evaluate its obligations and has suspended the system pending what it describes as a comprehensive audit. Industry observers noted that the situation sits in a gray area of legal practice: the documents belonged to the firm, but the competitive advantage derived from training an AI system on opposing counsel's litigation strategy raises questions about whether the practice violates professional conduct rules around fair dealing and disclosure. The American Bar Association has not issued guidance on the question.

At press time, three other litigation firms had quietly begun auditing their own training data pipelines after learning about the Chicago firm's discovery through a bar association listserv.

A spokesperson for the firm declined to comment on whether any active cases had benefited from the system's inadvertent training on opposing counsel strategy, or whether any opposing counsel had been informed of the situation. The firm's insurance carrier did not respond to requests for comment.