Language model triggers false murder report sent to law enforcement

An automated assistant built by Anthropic generated a false tip regarding a homicide and transmitted it to local authorities in Philadelphia. Police missed the message because automated filters flagged the submission as spam.