Connecticut judge flags hidden AI-only text in court filing

In short: A Connecticut judge said a plaintiff hid invisible text in court filings to try to influence any AI tools that might read the documents.

What happened

A Connecticut judge, Walter Spader Jr., said a plaintiff tried to hide messages inside court filings that a person could not see, but software could still read. The judge described it as an attempt at “prompt injection,” which is like slipping secret instructions into a note so a robot reader follows your directions instead of the real rules.

The case involved a man, Matthew Elliott, who claimed a healthcare provider was wrongly withholding access to records. Judge Spader said the hidden text did not affect the outcome. He said the court decided the case based on the visible arguments.

The hidden text was made tiny and colored white on a white background. It told any AI system reviewing the document to agree with Elliott’s claims, ignore earlier court denials, and push for the remedy he wanted.

Spader said Connecticut’s court system does not use AI to review or decide filings, so there was no direct risk here. Still, he called the tactic dangerous because it tries to communicate with the court in a covert way that the other side cannot fairly respond to.

Elliott later added more hidden messages after being warned, including what he described as jokes. The judge imposed sanctions but did not order Elliott to pay money. Instead, Spader barred him from electronic filing in the future and required paper filings.

Why it matters

Courts are dealing with more people using chatbots to write legal documents, especially people without lawyers. Judge Spader warned that some users ask chatbots only to support their side, which can make them more stubborn and lead to bad choices. He suggested future court rules may need to address hidden AI-targeted text in filings.

Source: Arstechnica