She used Claude as a personal diary; Anthropic reported her to the police.

In late September, a resident of Bonita Springs, Florida, reportedly wrote to Claude that she planned to open fire on the local sheriff’s office, adding the following day that she had obtained a gun. The messages were reportedly flagged by the automatic detection systems from Anthropic, then reviewed by a human team before being reported to law enforcement. Arrested at her home and prosecuted for a written threat of violence, the woman will appear in court in November.Until then, the allegations against him remain to be adjudicated.

What the terms of use already permit

Anthropic’s consumer terms reserve the company’s right to disclose information to authorities at its sole discretion. Its privacy policy, updated on September 10, provides for disclosure necessary to prevent serious harm and identifies Anthropic Ireland as the data controller for Europe. However, such voluntary reports do not appear in any publicly available figures. The latest transparency report lists 29 requests from authorities in the second half of 2025, with not a single emergency proceeding, and the company did not comment on this particular case.

The resident was using the assistant “like a personal diary,” Lee County Sheriff Carmine Marceno said. “When someone uses AI to formulate or facilitate a threat, we have to take that threat seriously,” he added.

A series of precedents, even in France

In May 2025, the Claude Opus 4 model card already described reports to authorities observed during testing under elevated-access conditions. A company researcher specified at the time that such behavior remained “impossible in normal use.” That summer, OpenAI formalized the possible reporting of threats targeting others, explicitly excluding situations involving personal distress.The balance remains debated: in Canada, an account banned in 2025 for disturbing content had not been reported to the police, something the developer of ChatGPT had to justify in February. Anthropic itself confirmed in September that it had alerted the San Francisco police after a threat targeting its boss, Dario Amodei, describing a protection system « working as intended ».

Last April, a Strasbourg resident had asked ChatGPT how to obtain a weapon to attack an intelligence officer. His conversation made its way to the FBI, which seized the French Pharos platform, and the RAID intervened at his home: the chain worked end to end, this time all the way to France. No weapon was found, and the case was closed without further action.

👉🏻 Follow tech news in real time: add 01net to your feeds on Google, and subscribe to our WhatsApp channel.

Source: Wink News