A man representing himself in a Connecticut lawsuit hid secret instructions inside his court filings, hoping any artificial intelligence system reading them would side with him. The instructions were written in tiny white text, invisible on the printed page but readable by any software scanning the document. They told an AI reader to agree with his arguments, ignore the court's earlier denials, and grant him the outcome he wanted.
Judge Walter Spader Jr. caught it. Court staff noticed odd spacing when the pages were printed, which led to the discovery. Spader ruled the hidden text had no effect on the case, a dispute involving a healthcare provider, but sanctioned the plaintiff anyway. He is barred from filing documents electronically for now and must submit paper copies in person.
This appears to be the first known case in the United States of someone trying this trick in a court filing, though a similar attempt reportedly surfaced earlier in Brazil. In later filings, the same plaintiff kept leaving hidden messages, some clearly just testing whether anyone was paying attention. The judge was not amused, calling any secret communication with the court a problem regardless of how serious the intent.
The pattern is bigger than one lawsuit. Researchers found the same tactic in academic publishing last year: hidden text reading, in effect, "give a positive review," planted inside dozens of research papers to fool AI tools that journals now use to help screen submissions. A separate study of two hundred thousand real job resumes found that at least one percent contained hidden text aimed at tricking the AI programs that many companies now use to filter applicants, and researchers say that share has climbed sharply since AI screening tools became common.
Three very different worlds, one shared trick: courts, academic journals, and hiring. In each one, a human used to read every document from start to finish. Now an AI often does the first pass, and people have figured out they can whisper instructions directly to the machine instead of persuading the human reader.
This matters for any business that has adopted AI to sort incoming paperwork, whether that is resumes, insurance claims, vendor proposals, or warranty complaints. A hidden line of white text is trivial to insert and easy to miss if nobody is looking for it. Courts are already further along this path than most companies realize: a recent survey found that more than sixty percent of federal judges now use some AI tool in their own work, even though many court systems, including Connecticut's, have no official policy allowing it. Policy and practice can drift apart quietly, and that gap is exactly where this kind of manipulation lives.
The fix does not require new technology, just a habit change. Anyone reviewing incoming documents with an AI tool should treat hidden or oddly formatted text the same way they treat a forged signature: a red flag for a human to check, not something the AI should quietly absorb and act on. The plaintiff in Connecticut got caught because a clerk noticed strange spacing while printing a page. Most companies do not have a clerk checking print spacing, which is precisely the gap this new trick is built to exploit.