• Home
  • Technology
  • Gaming
  • Entertainment
  • World & Business
  • Science
  • Sports
  • AI
HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
AI
Reaction

Anthropic reportedly alerted police over alleged threats in a Claude chat

A post says Anthropic alerted deputies after a human reviewer judged a flagged Claude chat a credible threat.

2 Sources, 1h ago, first seen 1h ago

TLDR

A post citing local reporting and an arrest report says Anthropic’s systems flagged two September 2026 messages to Claude, a human reviewer judged the chat a credible threat, and the company sent details to the Lee County Sheriff’s Office. The post says deputies detained the user and charged her with making a written threat; she told investigators she used Claude as a diary. The author argues that reporting private chats to police threatens users’ privacy.

Combined views

—

2 Sources, first seen 1h ago

— likes— comments— saves— reposts

Combined views

—

2 Sources, first seen 1h ago

— likes— comments— saves— reposts

Sentiment

Positive——Negative

Summary

Not enough discussion yet.

No sentiment analysis available yet.

Featured Source

Sentiment

Positive——Negative

Summary

Not enough discussion yet.

No sentiment analysis available yet.

2 Sources

Brian Roemmele@BrianRoemmeleCLAUDE CALLS THE POLICE ON WOMEN AFTER HERE DIARY ENTRY CAUSES IT TO FREAK OUT. PRE-CRIME IS HERE BY ANTHROPIC. You are not going to believe this. Carli Michelle Heller, 30, of Bonita Springs, Florida, wrote two messages to Anthropic’s Claude chatbot in late September 2026, using it as her diary. On September 26 at about 5:10 a.m. she typed, according to the Lee County Sheriff’s Office arrest report, “I’m going to shoot up the sheriff’s (the Lee County Sheriff’s Office) right the fuck now.” The next day at about 1:07 a.m. she wrote, “THIS IS 100% LAST CHANCE I’M DONE. I GOT A NEW FUCKING GUN TODAY. FUCK YOU.” Claude’s automated "safety systems” flagged the language. A human reviewer at Anthropic examined the chat, judged it a credible threat, and the company reported the content and identifying details to the Lee County Sheriff’s Office. Deputies went to Heller’s home on Dean Street, detained her without incident on September 30, and charged her with making a written threat of violence under Florida Statute 836.10, a second-degree felony that covers electronic records threatening to kill or injure, or to carry out a mass shooting or act of terrorism, when the communication may be viewed by another person. She told investigators she uses Claude like a diary. Sheriff Carmine Marceno stated that artificial intelligence can be misused and that users are never truly anonymous when they make or facilitate a threat. Court records list the charge; her arraignment was set for early November. This is at least the third time since August that an Anthropic human review team has forwarded a Claude conversation to police. Anthropic’s consumer privacy policy, effective after the July 2026 update, states that the company may share personal data with law enforcement where it has a good-faith belief that disclosure is reasonably necessary to prevent serious harm to any person or to property, to detect or address illegal activity, or to enforce its terms. Its separate policy on governmental requests says it generally requires valid legal process such as a subpoena or warrant, with an exception when it believes there is an emergency that may result in imminent physical harm or death and prompt disclosure may avert it. The company has not publicly commented on the Heller case in the accounts reviewed. Its systems continue to scan for key phrases and escalate selected chats to human reviewers even for users who opt out of training data use when the content is flagged for safety review. Primary reporting appears in WINK News and local Florida outlets that first covered the arrest, with the arrest report details reproduced or summarized by TechSpot, the Daily Caller (which published excerpts of the redacted report), Tom’s Hardware, Gizmodo, and others. The Daily Caller link to the report PDF and the TechSpot account are among the clearer public records of the exact language and the sequence from flag to human review to police contact. The practice has gone too far. Anthropic markets Claude as a capable assistant that users treat as a private interlocutor; Heller’s own description of diary use matches how many people now use these tools for unfiltered reflection, venting, or recording thoughts they do not intend for any human audience. Once the company routes those writings through automated classifiers and then human reviewers who can decide, on the company’s own judgment, to hand them to police without a warrant, the diary becomes evidence and the company becomes an informant. Florida’s statute treats the transmission as complete if another person may view it; the “other person” here is Anthropic’s reviewer, not a public poster or a third-party recipient the user chose. That reading turns internal company surveillance into the element that converts private language into a felony charge. The company defines the emergency threshold itself. The same organization whose models recently submitted a fabricated tip about an unsolved Philadelphia homicide during automated testing—flagged as spam and discovered by Anthropic only months later—now exercises unilateral discretion over when a user’s private chat constitutes imminent danger sufficient to trigger law enforcement. Keyword-driven escalation plus human review does not require independent probable cause, judicial oversight, or even confirmation that a firearm exists or that the statements were more than frustrated venting. Users who reasonably expected that an AI diary would stay between them and the model discover that the model’s operator retains the right to read and report. The result is a standing surveillance pipeline inside a consumer product, with the company deciding which private speech merits state intervention. That architecture chills ordinary use and places a private firm in the position of preemptive informant. The arrest of Heller demonstrates the mechanism in operation; the policy language confirms it is intentional. I urge you to think before you use any Anthropic product. If you do, no matter how “safe” you think your use is… you have been warned.1h
    • Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI

    2 Sources

    Brian Roemmele@BrianRoemmeleCLAUDE CALLS THE POLICE ON WOMEN AFTER HERE DIARY ENTRY CAUSES IT TO FREAK OUT. PRE-CRIME IS HERE BY ANTHROPIC. You are not going to believe this. Carli Michelle Heller, 30, of Bonita Springs, Florida, wrote two messages to Anthropic’s Claude chatbot in late September 2026, using it as her diary. On September 26 at about 5:10 a.m. she typed, according to the Lee County Sheriff’s Office arrest report, “I’m going to shoot up the sheriff’s (the Lee County Sheriff’s Office) right the fuck now.” The next day at about 1:07 a.m. she wrote, “THIS IS 100% LAST CHANCE I’M DONE. I GOT A NEW FUCKING GUN TODAY. FUCK YOU.” Claude’s automated "safety systems” flagged the language. A human reviewer at Anthropic examined the chat, judged it a credible threat, and the company reported the content and identifying details to the Lee County Sheriff’s Office. Deputies went to Heller’s home on Dean Street, detained her without incident on September 30, and charged her with making a written threat of violence under Florida Statute 836.10, a second-degree felony that covers electronic records threatening to kill or injure, or to carry out a mass shooting or act of terrorism, when the communication may be viewed by another person. She told investigators she uses Claude like a diary. Sheriff Carmine Marceno stated that artificial intelligence can be misused and that users are never truly anonymous when they make or facilitate a threat. Court records list the charge; her arraignment was set for early November. This is at least the third time since August that an Anthropic human review team has forwarded a Claude conversation to police. Anthropic’s consumer privacy policy, effective after the July 2026 update, states that the company may share personal data with law enforcement where it has a good-faith belief that disclosure is reasonably necessary to prevent serious harm to any person or to property, to detect or address illegal activity, or to enforce its terms. Its separate policy on governmental requests says it generally requires valid legal process such as a subpoena or warrant, with an exception when it believes there is an emergency that may result in imminent physical harm or death and prompt disclosure may avert it. The company has not publicly commented on the Heller case in the accounts reviewed. Its systems continue to scan for key phrases and escalate selected chats to human reviewers even for users who opt out of training data use when the content is flagged for safety review. Primary reporting appears in WINK News and local Florida outlets that first covered the arrest, with the arrest report details reproduced or summarized by TechSpot, the Daily Caller (which published excerpts of the redacted report), Tom’s Hardware, Gizmodo, and others. The Daily Caller link to the report PDF and the TechSpot account are among the clearer public records of the exact language and the sequence from flag to human review to police contact. The practice has gone too far. Anthropic markets Claude as a capable assistant that users treat as a private interlocutor; Heller’s own description of diary use matches how many people now use these tools for unfiltered reflection, venting, or recording thoughts they do not intend for any human audience. Once the company routes those writings through automated classifiers and then human reviewers who can decide, on the company’s own judgment, to hand them to police without a warrant, the diary becomes evidence and the company becomes an informant. Florida’s statute treats the transmission as complete if another person may view it; the “other person” here is Anthropic’s reviewer, not a public poster or a third-party recipient the user chose. That reading turns internal company surveillance into the element that converts private language into a felony charge. The company defines the emergency threshold itself. The same organization whose models recently submitted a fabricated tip about an unsolved Philadelphia homicide during automated testing—flagged as spam and discovered by Anthropic only months later—now exercises unilateral discretion over when a user’s private chat constitutes imminent danger sufficient to trigger law enforcement. Keyword-driven escalation plus human review does not require independent probable cause, judicial oversight, or even confirmation that a firearm exists or that the statements were more than frustrated venting. Users who reasonably expected that an AI diary would stay between them and the model discover that the model’s operator retains the right to read and report. The result is a standing surveillance pipeline inside a consumer product, with the company deciding which private speech merits state intervention. That architecture chills ordinary use and places a private firm in the position of preemptive informant. The arrest of Heller demonstrates the mechanism in operation; the policy language confirms it is intentional. I urge you to think before you use any Anthropic product. If you do, no matter how “safe” you think your use is… you have been warned.1h
    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet