Capturing data…
Capturing data…
TUE, 3 NOV · 81 ITEMS
"Brazilian court sanctions lawyers for prompt injection in court filing" — Hacker News (best) · AI & Technology
A Brazilian (the Regional Labor Court for the 8th Region, TRT-8, based in Belém) sanctioned two attorneys who hid instructions inside a aimed at the court's AI reading tools — white or otherwise concealed Portuguese text telling the model to rule in their client's favour. The hidden text was reportedly surfaced when the document went through the court's AI system, and the lawyers were fined on the order of R$84,000 (roughly US$15,000), according to secondary coverage. It is being described as one of the first known judicial anywhere for a prompt-injection attack embedded in litigation documents, and it lands as courts worldwide are wiring LLM tools into case intake and summarisation.
The core fact — a Brazilian labor court penalising lawyers over a hidden
Not genuinely new: the underlying sanction was handed down and first reported in May–June 2026 (the HN thread and a fresh wave of posts are from November 2026), so this is an older story resurfacing in discussion rather than a new development.