Man hides AI prompts in court filings in first known case of prompt injection attack

A Connecticut judge identified what appears to be the first documented case in the United States of a plaintiff attempting to hide prompt instructions in legal documents to manipulate AI analysis. The hidden text, formatted to remain invisible to human readers while legible to software, instructed AI systems to ignore prior court rulings and favor the plaintiff's arguments. Judge Walter Spader Jr. confirmed the hidden text did not influence the case outcome, but warned the tactic sets a dangerous precedent likely to be repeated as AI tools become more commonplace in legal systems.