Connecticut Judge Bars Self-Represented Litigant for Embedding AI Prompt in Filing
A Connecticut Superior Court judge found that a pro se plaintiff concealed AI-directed instructions in white-on-white text within multiple filings and revoked the plaintiff’s electronic filing privileges.
In Elliott v. N.Y. Bariatric Group, LLC, Judge Walter M. Spader, Jr. discovered that the plaintiff’s July 24 motion contained invisible white text formatted to be read by software but not by humans, directing AI tools to align their output with the plaintiff’s arguments and to overturn a prior clerk’s denial. Additional filings on the same day and later included abbreviated versions of the hidden instruction, as well as unrelated jokes and a concealed link to a Nosferatu video.
After a hearing on July 31 regarding possible sanctions, the plaintiff claimed the initial instruction was a good-faith audit of the court’s AI use, but continued the practice as a joke. The judge emphasized that such “prompt injection” undermines the open nature of legal communications and ordered the plaintiff’s e-filing rights rescinded, permitting only in-person paper submissions. The decision also referenced a similar sanction in Brazil’s labor courts, underscoring the emerging legal challenge of hidden AI directives.
Why it matters
It highlights how hidden AI commands can threaten the fairness of legal proceedings and prompts courts to set new rules.
In this story
