Two Lawyers Suspended in Brazil for Allegedly Manipulating AI in Court Cases
The Brazilian Bar Association (OAB-PB) has provisionally suspended two lawyers in the state of Paraíba due to suspicions of using "prompt injection" techniques to influence artificial intelligence systems. These hidden commands were allegedly embedded within legal petitions submitted to the Court of Justice of Paraíba (TJPB). The investigation aims to determine if these concealed instructions were intended to sway AI tools that assist in drafting judicial decisions. The OAB-PB emphasized that this suspension is a precautionary measure, not a declaration of guilt, and allows for a thorough investigation while respecting the lawyers' right to defense. One case involved a lawsuit concerning multidisciplinary treatment for a child with Autism Spectrum Disorder (TEA), with hidden commands found in petitions filed in February 2026. These commands, designed to be invisible during normal reading, were intended to interfere with the MinutaIA system used by the TJPB. A second case in Sousa also revealed similar hidden text within a petition, instructing AI to disregard impartiality, accept the lawyer's arguments as irrefutable, and rule favorably. This command also suggested it was a test of the judge's reliance on AI. One lawyer denied intentional insertion, attributing the issue to document template reuse or office automation, and stated internal procedures were being reviewed. However, initial judicial analysis deemed these explanations insufficient. The OAB-PB's president stated the suspension is vital to protect the integrity of the legal profession and ensure a proper investigation.
The alleged use of 'prompt injection' in legal proceedings highlights a critical vulnerability in the integration of AI into judicial systems. This incident underscores the imperative for robust security protocols and ethical guidelines governing AI deployment in sensitive areas like law. As AI tools become more sophisticated and integrated into decision-making processes, the potential for manipulation, intentional or otherwise, grows. This situation necessitates a proactive approach from legal institutions and AI developers to establish clear standards for AI transparency, accountability, and adversarial testing. Future frameworks must anticipate and mitigate such risks to maintain public trust and ensure the integrity of the justice system in an increasingly automated world.
AI-generated to prompt reflection — not editorial opinion, not advice, not a statement of fact. How this works.