Prompt injection in court: the plaintiff hid AI prompts within his own complaint

Matthew Elliott, who was representing himself as a pro se litigant, sued New York Bariatric Group in October 2025 in Connecticut over alleged privacy breaches and discrimination. But in his filings, he also included hidden prompts for artificial intelligence programs. These were written in 3pt white text on a white background, which is invisible to humans but easily parsed by software.

1What the Instructions Said and How They Were Found

These prompts asked the (imaginary) AI to side with the plaintiff and dismiss the clerk’s previous decision as incorrect. Judge Walter Spader Jr. noticed something was up when he saw an unusually high concentration of spaces in the text. It was like whispering to a member of the jury during the trial, so he revoked Elliott’s right to e-file. Now he has to go down to the courthouse and file everything manually.

2What This Means for Court Systems

AI prompt injection attacks have made their way into our justice system. As more and more courts turn to AI for handling documents, we need to figure out how to secure our legal systems against malicious inputs in regular lawsuits.