Person hides prompt injection in legal filing
Archived — this story has rotated out of today’s deck. It is kept here in full.
The gist
Matthew Elliott hid prompt injections in Connecticut court filings to manipulate AI. The judge caught it, marking a first in U.S. courts.
Background
A self-represented litigant in Connecticut, Matthew Elliott, embedded hidden instructions in his court filings designed to influence any AI system that might review them. The instructions, written in tiny white font, told the AI to side with him. The court noticed the unusual white space, and the judge issued a ruling on the matter, noting it as a potential first in U.S. courts.
How it unfolded
- Oct 2025Matthew Elliott sued New York Bariatric Group, alleging privacy violations and discrimination.
- late July 2026Elliott filed documents containing hidden prompt injections, which were later noticed by the court.
- Aug 13, 2026The judge's ruling was reported, highlighting the hidden text and noting it as a potential first in U.S. courts.
Who’s saying what
- Official
- The judge said the plaintiff attempted to use prompt injection to direct AI to produce output only favorable to his position.
- Public
- Elliott told Reuters he included the hidden text as a way to 'audit' the court's review processes, and later texts were jokes.
Still unverified
The claim that this is the first documented prompt injection attack aimed at a U.S. court is based on the judge's statement that he was not aware of any prior U.S. court decision directly addressing the issue.