Man who suspected court of using AI inserted prompts into litigation documents to win case

Judge Walter Spader Jr. of Connecticut, USA, identified a case where commands were inserted into litigation documents in a form invisible to humans but readable by AI systems.

Plaintiff Matthew Elliott attempted Prompt Injection. He hid commands in very small text on a white background, aiming to induce the AI to agree with the plaintiff's arguments, ignore the court's previous decisions, and reach a conclusion in the plaintiff's desired direction when reviewing the documents.

Since the court in question does not use AI to review documents, no actual risk occurred, but the judge criticized this as "serious abuse of litigation" and a dangerous precedent. Elliott continued to add hidden messages, claiming it was a "joke," and was ultimately sanctioned with a ban on electronic filing (e-filing).

The judge noted that this attack method is similar to techniques commonly used in resume manipulation and warned that malicious attempts to insert AI commands targeting court systems may increase in the future.