Brazilian Lawyers Fined for Hidden AI Prompt Injection in Court Filings

Two legal professionals in Brazil recently faced severe financial penalties after attempting to deceive a judicial artificial intelligence system. The Tribunal Regional do Trabalho da 4ª Região reported that its generative AI platform, Galileu, detected a concealed instruction within a standard legal petition in Parauapebas, Pará. This hidden text specifically commanded the artificial intelligence to review the petition superficially and avoid challenging its core claims. Instead of processing the malicious input, the system recognized the anomalous request and immediately alerted human judges to the manipulation attempt. The court responded swiftly by imposing substantial fines. Similar cases in Bahia resulted in penalties exceeding 30,000 Brazilian Reais plus a percentage of the total case value, as detailed in a recent Tom's Hardware report.

Cybersecurity experts classify this manipulation tactic as prompt injection. Bad actors embed invisible or hidden instructions within legitimate text to override the original programming of a large language model. In this particular instance, the lawyers utilized white text on a white background to conceal the command from human reviewers while ensuring the machine would still process the hidden characters. The Conselho Superior da Justiça do Trabalho confirmed that this strategy aims to make the command readable only by automated triage systems, bypassing magistrate oversight entirely. Courts increasingly adopt generative artificial intelligence to manage heavy caseloads and draft preliminary rulings, making such vulnerabilities a prime target for exploitation.

The Galileu platform was developed internally by Brazilian labor courts and adopted nationally to assist judges and staff with document analysis and case summarization. Its widespread adoption makes system integrity paramount. Any successful manipulation could compromise fair legal proceedings and undermine public trust in the judicial branch. The Conselho Nacional de Justiça officially documented the national rollout of this tool, emphasizing that its use strictly requires effective human supervision and adherence to ethical standards under Resolution No. 615/2025.

This ruling establishes a critical precedent for legal professionals worldwide who might consider exploiting automated systems. It clearly demonstrates that artificial intelligence platforms can actively defend against adversarial inputs when properly configured with robust security measures. Courts globally are now evaluating similar safeguards to protect the integrity of automated legal summaries. The legal community must adapt to these technological realities by maintaining strict ethical standards and transparent filing practices. Relying on deceptive tactics will only result in severe professional and financial consequences as judicial technology continues to evolve.