A Connecticut judge has documented what appears to be the first known attempt in a US court to smuggle invisible prompts into a legal filing, hoping an AI system would read instructions a human couldn’t see. Judge Walter Spader Jr. confirmed the hidden text played no role in the outcome of Matthew Elliott’s case against a healthcare provider over access to records. But the tactic itself, Spader said, sets a “dangerous” precedent as courts increasingly experiment with AI tools.
The hidden instructions were shrunk to tiny-point type and colored white on a white background — invisible to the naked eye but perfectly legible to any software parsing the document’s text. The prompts directed any AI reviewing the filing to agree with Elliott’s arguments, ignore the court’s prior denials, and ensure remediation unfolded exactly as he wanted. Classic prompt injection, dressed up in legal formatting. Elliott seemingly hoped to revive arguments that had already been defeated on the merits.
He did not get away with it. The court weighed the filing normally, and Elliott still faced sanctions — modest ones, but sanctions nonetheless — because he kept adding hidden text even after being warned. His later efforts included a link to a Nosferatu YouTube video, a message reading “hi :) I hope yo ucant see me,” and a garbled line about sending “RE GARBS.” Elliott called these jokes. The court called the overall pattern “serious litigation abuse.”
This is almost certainly not the last time. Courts nationwide are piloting AI summarization and document-review systems, and litigants who understand how those models work will test whether invisible instructions can nudge outcomes. The Connecticut case offers a clean template for any judge who encounters this next: review the filing on its merits, then sanction the deception separately. The harder question is what happens when the hidden text isn’t a joke and the AI tool isn’t carefully monitored. That’s the scenario keeping court administrators up at night.
