Source: Tom’s Hardware

A Brazilian labor court fined two lawyers after finding that the initial complaint they signed contained concealed instructions intended to influence the court’s AI system. Professor Vladimir Aras, in a post reproduced in the report, places the proceeding in Pará, within the Eighth Labor Court Region, and identifies the system as Galileu. According to the machine-translated decision quoted in the report, the court discovered white lettering against a white background while processing the complaint through the system. Changing the lettering’s color revealed a command directing the AI to respond superficially to the complaint and leave the documents unchallenged, regardless of other instructions. The decision described the technique as prompt injection.

The court assigned responsibility exclusively to the two lawyers who signed the complaint, reasoning that the claimant lacked access to the document and sufficient knowledge of legal technicalities to have altered it. According to the report, the court regarded the conduct as offensive to the dignity of justice and imposed financial penalties. The lawyers were also reported to the state bar association and the Regional Labor Court’s Internal Affairs Department. The court said that the defendant had failed to respond and that the claimant therefore prevailed by default; the report consequently described the attempted AI manipulation as unnecessary to that outcome.