A Brazilian court has sanctioned a group of lawyers after discovering that their court filing contained hidden instructions designed to manipulate an artificial intelligence system. The embedded text, a form of prompt injection, was intended to steer how an AI tool processed or summarized the legal document. Instead of treating the tactic as harmless experimentation, the court ruled that it crossed a line into improper conduct and imposed penalties on the attorneys involved.
Prompt injection has become a familiar problem in AI systems, where hidden commands can override a model's intended behavior. What makes this case notable is that the injection appeared in an official legal proceeding rather than a security test or research setting. The lawyers apparently believed the hidden text would go unnoticed, but the court examined the filing closely enough to identify the manipulation attempt. The sanctions reflect growing awareness among judges and regulators that AI tools are now part of legal workflows and can be abused.
The ruling raises broader questions about how professional rules apply when humans and AI systems interact in formal settings. Legal ethics codes already prohibit misleading the court, and the Brazilian decision suggests that deceptive instructions aimed at a machine can fall under the same principle. Observers say the case could encourage other courts to scrutinize filings for hidden AI directives and to set explicit rules for how attorneys may use generative tools.
For law firms, the episode is a warning that AI adoption requires new internal controls. Attorneys may need to review documents for concealed text, document how AI tools were used, and train staff on the risks of prompt injection. The Brazilian court's sanctions are unlikely to be the last word on the subject, but they establish an early precedent that manipulating AI in litigation carries real professional consequences.