Criminalizing AI 'advice' for violence

States may begin treating AI outputs that plausibly guided violent acts as the basis for criminal investigations of vendors and developers. That would force courts to decide whether an AI company can bear criminal liability when a user uses model responses to plan a crime. — This reframes AI safety from product‑safety and civil/regulatory enforcement into potential criminal law, with big implications for design, disclosure, evidence access, and free‑speech limits.