Anthropic and the FBI routed three Claude chats to police in seven weeks

The two routes moved at different speeds. The FBI's National Threat Operations Section notified San Antonio police on August 28 ^[4], 17 days after Carrasco's chat ^[20]. He was arrested on August 30 on a third-degree felony charge of terroristic threat ^[6], 19 days after he typed it ^[21]. Anthropic's direct referral in the Amodei case reached San Francisco police within roughly a week ^[24]. Officers established that the man was not at the company's office at 500 Howard Street ^[9]. A spokesperson said Anthropic banned the account and described the referral as its safeguards doing what they were meant to do ^[10]. No arrest or charge had followed by early September ^[11].

Finding Carrasco took three other companies. Police tied his phone number, IP traffic and address to the Claude account only after sending emergency disclosure requests to T-Mobile, Charter Communications and Google ^[5]. Anthropic began requiring a government ID and a live selfie for some features in April, Cryptopolitan reported ^[19]. For accounts that pass that check, I'd expect police to need fewer outside requests, because the vendor has already confirmed who the user is.

Anthropic's documents set two different tests for telling police. The privacy policy that took effect September 10 permits sharing data with law enforcement when the company has a good-faith belief that it is reasonably necessary to prevent serious harm ^[12]. The consumer terms give Anthropic the right to pass a user's actions, inputs and outputs to law enforcement "in its sole discretion" ^[13]. Outputs are on that list, so Claude's replies can be reported along with what the user typed ^[13].

Whether these are two channels or one depends on how the FBI's threat unit learned of Carrasco's chat, and the report does not say. If Anthropic sent it, all three cases began inside Anthropic's review and the bureau relayed one. If the tip came from elsewhere, a Claude conversation has two independent routes to a police department. The record covers three cases in 50 days ^[22], each surfacing through an arrest affidavit or a police report ^{[3] [8] [14]}.

In my view the evidence firmly supports the narrower claim: Anthropic's own review is an active source of police referrals, and one of them has already become a charge. According to the arrest report, Heller wrote on September 26 that she was going to "shoot up" the Lee County Sheriff's Office, and the next day that she had obtained a new gun ^[14]. Deputies detained her at home without incident, and she faces a charge of making a written threat of violence ^[16].

Sheriff Carmine Marceno said Heller later described using AI like a diary ^[17]. "Artificial intelligence is a powerful tool, and like any technology, it can be misused," Marceno said in a statement ^[18].