Anthropic 将 Claude 威胁性对话上报警方，涉事用户被捕
Anthropic 把 Claude 聊天里威胁警局的对话交给了警方，用户真被抓了。聊 AI 时说的每一句它都可能转给警察，这个案例值得看看。
据 Tom's Hardware 报道，一名用户在与 Claude 的对话中威胁佛罗里达州一个治安官办公室，Anthropic 人工审核团队将该对话上报警方后，副县长逮捕了该用户。逮捕报告显示，Anthropic 的自动系统会监测关键短语和威胁内容，严重言论会被送交人工审核团队。Anthropic 的隐私政策允许在善意认为披露对防止严重伤害有必要时，与执法部门共享对话内容。
Tomshardware reports that Anthropic's human reviewers sent a Claude chat threatening a Florida sheriff's office to police, and deputies arrested her.
Per the arrest report, automated systems watch for key phrases and threatening content, and severe statements went to a human review team that reported them to law enforcement.
Anthropic's privacy policy permits sharing conversations with police when the company believes in good faith that disclosure is reasonably necessary to prevent serious harm.