Anthropic reported diary entry to police, woman faces felony charge

Original Article ↗ Hacker News Discussion ↗

A Florida woman was charged under a state law on written threats after Anthropic’s Claude chatbot flagged and forwarded diary-like messages in which she said she would “shoot up” the local sheriff’s office. Commenters debate whether such private-seeming AI chats meet the legal standard of a “communication … in a manner in which it may be viewed by another person,” raising concerns about thought crime, intent, and constitutional limits on speech. The incident becomes a broader proxy for worries about AI services as mass-surveillance tools, unclear user expectations around privacy, and pressure on tech firms that can be blamed both for reporting and for failing to report users’ violent fantasies.

Legal interpretation of the threat

  • Core statute: Florida law makes it a felony to send or transmit written threats in any medium “in a manner in which it may be viewed by another person.”
  • Major dispute: whether entering text into Claude as a “private diary” counts as such transmission.
  • One side argues:
    • The message was in fact viewed by Anthropic staff, so the statutory condition is met.
    • Laws about threats and conspiracy often hinge on actions plus words, not just intent.
  • Other side argues:
    • The law targets threats communicated to people, not private notes or journaling.
    • There may be no “mens rea” if the user reasonably believed no human would read it.
    • Analogies are raised: handwritten diaries, email drafts, Google Docs, cloud backups, etc.

Expectation of privacy with LLMs

  • Many commenters say non-technical users reasonably assume chats with AI are private, akin to diaries or personal notes.
  • Others insist:
    • “Not your computer, not your data” – anything sent to a cloud service should be assumed readable and retainable.
    • Anthropic’s policies explicitly allow human review of flagged content, though few users read these terms.
  • Some call this “spying with extra steps”; others say it’s simply using a commercial service under a disclosed EULA.

Anthropic’s decision to report

  • One camp sees reporting as responsible and legally necessary, especially after prior criticism of AI labs for not flagging violent plans.
  • Another camp argues Anthropic created the surveillance problem by auto-screening all chats; without that, there would be nothing to report.
  • Several frame this as “damned if you do, damned if you don’t” PR:
    • Either “AI ignored shooter warnings” or “AI snitches on private thoughts.”

Thought crime, safety, and broader implications

  • Strong concern about “Minority Report”–style pre-crime and the erosion of spaces for unfiltered private thought.
  • Counterpoint: some want systems that detect unstable or violent intent early to prevent shootings or suicides, comparing AI to therapists with mandatory-reporting duties.
  • Repeated calls for:
    • Clearer disclosures in AI UIs about monitoring and potential reporting.
    • Legal protections for “private mode” or end-to-end–encrypted AI use.
    • Greater use of local/open models for genuinely private journaling and experimentation.