A Connecticut court filing included hidden prompt instructions meant to influence a possible AI review system, according to Ars Technica. The court said it does not use AI to review filings, but the judge still treated the attempt as serious misconduct.
The tactic relied on text that humans were unlikely to notice, a familiar prompt-injection pattern from software security applied to the legal process. The goal was not to persuade a person directly, but to exploit an imagined automated reader.
For courts and law firms, the case is a warning that AI risk does not begin only after institutions deploy a chatbot. People may already act as if automated review exists, and they may try to manipulate it through filings, exhibits or metadata.
The immediate sanction was procedural: the litigant lost electronic filing privileges. The broader issue is governance. Legal systems will need rules for hidden machine-targeted text before AI tools become routine in document review.