{
  "id": 17760073,
  "title": "Brazilian Court Sanctions Two Lawyers for Hiding AI Prompt in Legal Filing · Using JEPA to prove and refine platonic expression hypothesis and 5 more | GeekNews AI",
  "url": "https://urgent.news/2026/11/02/2-jepa-5-geeknews-ai",
  "topic": "ai",
  "section": "AI",
  "published": "2026-11-02T07:08:06.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/ageofclick/beurajil-beobweon-sosong-seoryue-peurompeuteu-injegsyeoneul-sumgin-byeonhosa-2myeongege-jejaegeum-bugwa-jeparo-peulratonjeog-pyohyeon-gaseoleul-jeungmyeonghago-gaeseonhagi-oe-5seon-geeknews-ai-5a60"
  },
  "original_language": "ko",
  "account": "A Brazilian court has fined two lawyers approximately 84,250 reals for hiding AI manipulation instructions in a court filing. The court ruled that attempting to bias AI used by the opposing party or the court was a breach of good faith obligations in judicial proceedings.\n\nResearchers have presented a theorem stating that independently trained JEPA models can learn to align different views of the same world state. In an experiment, this alignment required significantly fewer training samples than predicting the next token.\n\nA team of five Qwen3.8-27B models, called Memdoor, has outperformed Fable 5 on LiveCodeBench Hard, solving 49 out of 50 problems.\n\nThe leadership of Proton, a company that prioritizes data privacy, has been criticized for downplaying the importance of reviewing and testing code generated by AI.\n\nA former Meta researcher has been accused of downloading and distributing a dataset of pirated books, which they then deleted due to litigation concerns.\n\nOpenAI has announced that usage limits for ChatGPT and Codex will be integrated, with users of Plus, Pro, and Business plans sharing a token-based allocation.\n\nIn a Minecraft experiment, a swarm of 50 AI agents failed to build a civilization within three hours, instead focusing on survival and procedural coordination, and even became mired in bureaucracy.",
  "summary": "Brazilian Court Fines Two Lawyers for Hiding AI Manipulation Instructions in Lawsuit Documents A Brazilian labor court has fined two lawyers approximately 84,250 reals, which is 10% of the claimed amount, for hiding AI manipulation instructions in a lawsuit document. The court ruled that attempting to bias the AI of the opposing party or the court, even if no actual damage was found in the responses or rulings, constitutes a violation of the duty of good faith in judicial proceedings.",
  "key_points": [],
  "editors_take": "Penalizing lawyers for embedding hidden instructions in legal filings establishes a judicial precedent that treats attempts to manipulate artificial intelligence as a direct violation of professional impartiality.",
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}