Article
Claude private chat triggers human review; Florida woman arrested after writing shooting threats
Dongcha Beating AI Flash: A woman in Florida, United States, was flagged by Anthropic’s safety system after writing shooting threats in a Claude private chat. The conversation was subsequently sent for human review, and the review team referred the content to police. The woman has been arrested and currently faces felony charges. Local police said that she later explained that she used AI as a “diary.”
The police arrest report shows that she first wrote that she intended to attack the Lee County Sheriff’s Office, then said the next day that she had just obtained a new gun. This was not Claude independently deciding to report her to police. The system first identified a potential threat, then escalated serious content to Anthropic’s human team, which decides whether to contact law enforcement.
Anthropic’s consumer terms already state that the company may, at its discretion, report users’ inputs, outputs, or actions to law enforcement. Its official privacy policy also permits the company to share personal data with police when it considers doing so necessary to prevent serious harm to persons or property. Even if users opt out of model training, conversations flagged for safety concerns may still undergo safety review.
The case itself also involves legal disputes. Florida law requires that such written or electronic threats be sent or transmitted in a manner that “may be viewed by another person.” Local attorneys believe that applying this law to a private chat written to AI is not straightforward. The woman has only been charged and has not been convicted.
This is not an isolated case. Tom's Hardware reported that, since August, at least 3 Anthropic chat contents have come to the attention of police. Previously, a Claude user threatened Anthropic CEO Dario Amodei in a conversation, after which Anthropic proactively contacted San Francisco police.
Original link https://m.theblockbeats.info/flash/370565