STF identifica primeira tentativa de fraude processual com IA
2026-09-25
Correio Braziliense
The information displayed in the AIM should not be reported as representing the official views of the OECD or of its member countries.
The Brazilian Supreme Court (STF) detected a prompt injection attempt in a legal petition, where hidden commands aimed to manipulate its AI system, Maria Shield, for favorable judicial outcomes. The attempt was uncovered, did not affect decisions, and led to disciplinary and criminal investigations, including a fine for the responsible lawyer.[AI generated]
Why's our monitor labelling this an incident or hazard?
The event involves an AI system used for automated judicial analysis, which was deliberately targeted by a prompt injection attack to manipulate judicial decisions. This constitutes a misuse of an AI system in a legal context, aiming to influence judicial outcomes unfairly, which is a violation of legal rights and procedural fairness. Although the manipulation attempt was unsuccessful and caused no actual harm, the event describes a direct attempt to cause harm through AI system misuse. According to the definitions, an AI Incident includes events where AI system use or misuse has directly or indirectly led to harm or violations of rights. The attempt to manipulate judicial decisions via AI qualifies as an AI Incident due to the direct involvement of AI and the nature of the harm intended (violation of legal procedural rights).[AI generated]