Brazil cracks down on AI prompt‑injection and election‑rule breaches

Brazilian courts have identified the use of "prompt injection" – hidden commands embedded in electronic legal documents – to manipulate generative AI tools that assist judges. Judges in Minas Gerais imposed fines and disciplinary measures on lawyers and classified the practice as a new form of litigating in bad faith. The national judicial intelligence centre issued technical guidance on preventing and sanctioning the behaviour.

A separate study found that major AI services, including ChatGPT, Gemini, Meta, Grok and DeepSeek, are still ranking and recommending political candidates ahead of Brazil's August 16 election propaganda period, contrary to rules set by the Superior Electoral Court (TSE). The platforms often answer voter‑guidance queries without citing sources, raising concerns about undue influence on the electorate. The platforms have until August to adjust their operations to comply with TSE regulations.