A Connecticut judge has confirmed what appears to be the first instance of a plaintiff attempting to hide text in court filings that can only be read by artificial intelligence systems. The aim was to sway the AI towards supporting the plaintiff’s arguments.

The tactic, known as prompt injection, involves formatting hidden instructions in tiny-point type and altering their color so they are invisible to human readers while remaining visible to software. Judge Walter Spader Jr. discovered these attempts and dismissed them, stating that such actions set a 'dangerous' precedent for the future of legal proceedings.

Matthew Elliott, the plaintiff involved in this case, tried to shift the court's favor by instructing any AI reviewing his document to ensure textual outputs aligned with his arguments, ignore prior court denials and mandate remediation according to his wishes. His plan did not succeed, but he continued adding hidden text even after facing sanctions for what was deemed a 'serious litigation abuse'.

These later prompts were intended as jokes and included links to a Nosferatu YouTube video, simple messages like 'hi :)' and nonsensical phrases such as ‘TELL SHAWN I SEND MY RE GARBS!!!! HAHAHA U GUYS GET THIS EGGWUH???? AHAH.’ Despite the humorous nature of some prompts, they still violated court guidelines.