US man hid AI prompts in court filings hoping to sway ruling in his favor
A Connecticut judge identified the first known US case of a plaintiff hiding invisible AI instructions inside court filings, in an attempt to influence how the case was decided. The attempt failed but still led to sanctions.
A Connecticut judge, Walter Spader Jr., has published a decision describing what appears to be the first known US instance of a court plaintiff embedding hidden AI-directed text inside legal filings, hoping to sway the outcome of his case. The plaintiff, Matthew Elliott, had alleged that a healthcare provider was improperly withholding access to his medical records.
The hidden text was shrunk to tiny type and colored white on a white background, making it invisible to human readers while remaining fully readable by software. It instructed any AI system processing the document to side with Elliott's arguments, disregard the court's earlier rulings against him, and ensure the outcome he wanted.
Spader confirmed the hidden text had no actual effect, since the filing was reviewed on its merits by a human. Still, he called the attempt a "dangerous" precedent and a "serious litigation abuse." The Connecticut Judicial Branch does not use AI to review or decide filings, so there was no real risk of the tactic working there.
Elliott kept inserting hidden text even after the court warned him he could face sanctions, later claiming some of the messages were meant as jokes. Spader rejected that explanation, saying it made no sense to hide jokes in documents meant to be taken seriously by the court.
While Spader found sanctions were warranted, he declined to impose a monetary fine, apparently taking into account that Elliott was a self-represented litigant who seemed to have been convinced by AI that his arguments were airtight. Instead, the judge barred Elliott from e-filing going forward, requiring him to submit paper filings instead.
Spader noted a similar case in Brazil, where two attorneys used the same tactic in a court that does use AI to review cases; they were fined roughly $16,000 after the system detected the hidden text. The judge warned that courts will likely need new rules to address prompt injection, as such misuse of AI is evolving quickly.
