A plaintiff hid secret AI instructions in court filings. The judge wasn't a bot.

A litigant tried to win his lawsuit by hiding invisible instructions for AI software inside his court filings.

According to Ars Technica, Connecticut plaintiff Matthew Elliott shrank text to a tiny font and colored it white on a white background inside electronic legal briefs. Fearing the court used AI to summarize cases, Elliott embedded prompt injections — hidden text that instructs an AI how to act — telling software to agree with his arguments and ignore past court denials. He later hid jokes, including a link to a Nosferatu video. The snag: the Connecticut Judicial Branch does not use AI to review filings. Judge Walter Spader Jr. spotted the text and banned Elliott from electronic filing, forcing him to submit future papers by hand.

Why it matters: This appears to be the first documented prompt injection attack in a US court. Judges previously focused on keeping hallucinated AI quotes out of legal filings. Now, courts face the opposite problem: litigants trying to secretly manipulate the software reading their documents.

Know this: The tactic is failing elsewhere, too. When attorneys tried a similar trick in a Brazilian court that actually used AI, the software caught the hidden text before processing it, resulting in roughly $16,000 in sanctions.

When trying to convince a court, plain visible words still beat secret white text every time.