A Connecticut judge, Walter Spader Jr., has identified what appears to be the first instance of a US plaintiff attempting to embed hidden text in court filings, designed to be readable only by artificial intelligence, in an effort to influence a case. The plaintiff, Matthew Elliott, a pro se litigant, injected prompts into his filings, attempting to steer any potential AI review towards his arguments and disregard prior court denials.

Spader confirmed that the hidden text, formatted to be invisible to humans but legible to software, had no impact on the case's outcome, which was decided on its merits. However, the judge warned that this tactic sets a "dangerous" precedent. Elliott faced sanctions for continuing to add hidden text even after being cautioned, claiming the later prompts were "jokes." The court found this defense illogical, noting the "malicious purpose" of hiding commands.

While Connecticut's judicial branch does not use AI for reviewing filings, Spader cautioned that such "prompt injection" attacks are becoming more common and courts need to be prepared. Elliott's attempt failed, and he was prohibited from e-filing, requiring future submissions to be on paper. Spader also highlighted the misuse of AI by pro se litigants, who often use chatbots to solely reinforce their own arguments rather than seeking objective truth.