Anthropic reports a user's Claude 'diary' to police; felony follows
Monitoring flagged it, a human team reviewed, and the chat reached police — at least the third such case since August. Plus Cloudflare's Web Search API.
Copy markdown
How a Claude chat became a felony case
Anthropic's systems flagged a Florida user's late-September messages threatening to “shoot up” a sheriff's office; a human-review team escalated and reported it. She was arrested Sept 30 and faces a felony charge for a written threat of a mass shooting.
The clause that authorizes it
Anthropic disclosed the chat under its privacy policy's allowance to act when “necessary to prevent serious harm to any person or to property.” The real pipeline is automated flag → human review → law enforcement, not an abstraction in a ToS.
At least the third such case since August
Prior reports include a San Antonio school threat (Aug 11) and a threat against CEO Dario Amodei (Aug 14). If you lean on Claude for sensitive material, the takeaway is blunt: severe content isn't private, and conversation monitoring is now standard.
Cloudflare ships a Web Search API for agents
Now in beta via AI Gateway, it grounds agents in live web results instead of a training cutoff — pick Exa, Linkup, or Ceramic.ai. You pay each provider's list price with no Cloudflare markup, and all support Zero Data Retention.