A US judge has warned of a "dangerous" precedent after a litigant hid secret prompts inside court documents to manipulate artificial intelligence systems. Connecticut judge Walter Spader Jr said Matthew Elliott used tiny, white-on-white text designed to be invisible to humans but readable by software. The hidden instructions ordered any AI tool reviewing the filing to agree with Mr Elliott's arguments in a dispute over healthcare records. The attempted prompt injection failed to influence the court, which decided the case on its merits. However, Mr Elliott received modest sanctions after continuing to insert the hidden text despite prior warnings from the court.

Background: AI in Courts and Sentencing

← Back to the news feed