A former Goldman Sachs analyst wrote out a full plan to kill his ex-girlfriend in ChatGPT. OpenAI picked up on the dangerous signals through chat logs and decided to alert the authorities, preventing a tragedy. However, just six months ago, OpenAI let a flagged dangerous account slip through the cracks, which ultimately led to a tragedy. AI seems to be evolving from just a chat tool into something entirely new: it might hear a person’s most private—and most dangerous—thoughts. These conversations could one day become clues for solving cases or even prevent crimes before they happen. But some are raising questions: when AI starts acting as a 'digital witness,' should saving lives or protecting privacy take priority?