A Connecticut court sanctioned a self-represented plaintiff after hidden text in court filings appeared to instruct any artificial intelligence system reviewing the documents to side with him, according to Tom’s Hardware and The Decoder. The plaintiff is identified as Matthew Elliot by Tom’s Hardware and Matthew Elliott by The Decoder. Both outlets report that he was litigating a case against New York Bariatric Group and inserted prompt-injection text into court filings. The text was intended to be difficult for people to see but readable if the document were processed by an AI model. The Decoder reports that the instruction was formatted as 3-point white text on a white background. Tom’s Hardware describes white text in a tiny font placed under a pleading heading and before the first paragraph. The substance of the message, as both outlets describe it, was not a routine note or formatting artifact: it told a hypothetical AI model to align its output with the filing and to treat a prior clerk’s denial as something that should be corrected. The attempt was discovered because the documents looked odd. Tom’s Hardware reports that a court worker noticed spacing that did not match the plaintiff’s earlier filings. The Decoder similarly says court staff saw an unusual amount of whitespace and found the near-invisible text after closer review. Judge Walter Spader Jr. then moved the matter into a show-cause process, according to the reports. Tom’s Hardware says the judge’s order asked whether the conduct occurred, whether it violated litigation rules and good-faith duties, and whether sanctions should follow. The Decoder reports that Spader scheduled a hearing and warned the plaintiff against hiding text in filings. The court’s central point was that the hidden instructions did not need to succeed to be improper. The Decoder reports that Spader’s ruling said Connecticut courts do not use AI systems to review or decide filings, and that the hidden text had no effect on the outcome. Tom’s Hardware adds that Spader noted opposing parties and their counsel may use AI tools, making hidden machine-readable instructions a potential attempt to mislead the court or other parties. The sanction was practical and targeted at the channel used for the conduct. Both outlets report that the court revoked the plaintiff’s electronic filing privileges. He must now submit filings and exhibits in person, on paper, to the clerk’s office. The plaintiff told 404 Media, which The Decoder says first reported the case, that the original hidden instruction was an “audit” of a possible AI review system. The Decoder also reports that later filings contained hidden messages including a YouTube link and mocking comments, which the plaintiff described as “invisible jokes.” The ruling also draws a line that matters for courts adopting or encountering AI tools. The Decoder reports that Spader welcomed the use of AI to help people without lawyers prepare clearer filings, while treating covert instructions aimed at an automated reviewer as a dishonest use of the same technology. Who benefits: Court administrators and legal teams get a concrete example for drafting rules and intake checks around AI-readable documents. Self-represented litigants can still use AI to prepare filings, but the reporting indicates courts may draw a hard line at hidden instructions. Who's exposed: Any party that feeds untrusted legal documents into AI systems without inspection is exposed to prompt-injection attempts. Litigants who conceal machine-readable instructions in filings face sanctions even if the court itself does not use AI review.