The New Yorker tells how a little-known Meta team reports hundreds of threats each year to law enforcement, despite increasingly difficult codes to decipher.
The New Yorker devotes a long investigation to Meta's law enforcement liaison team, co-led by Emily Vacher, a former FBI officer specializing in crimes against children. Since 2018, this group has brought together former investigators and journalists who have studied the traces left online by past killers. With engineers, they built AI systems that detect suspicious signals; Each alert is then examined by a human, who looks at publications, exchanges and even requests addressed to chatbots. When "actual acts" (list of targets, manifesto) appear, the information is passed on to the police: Meta did so about 800 times in the world last year.
A coded language that constantly evolves
The article opens with the case of two teenage girls from Texas, who were arrested in February 2025 after Meta was reported to the FBI, and cites a Queens case where an armed student was arrested just over two hours after an Instagram publication. But most cases are less clear: the online community known as the True Crime Community, linked according to the Institute for Strategic Dialogue to at least nine attacks and sixteen projects foiled in two years, bypasses moderation with nicknames, childhood photos of killers or videos generated by IA. The ISD also blames the platforms for reducing their trust and security teams while amplifying these content with their algorithms.
The investigation finally recalls the opposite risk, that of under-signaling: in 2025, OpenAI had identified and closed the account of a teenage girl who evoked killing scenarios with ChatGPT, without warning the police; She then killed six people in Tumbler Ridge, British Columbia, and the company is now the subject of several complaints.
Read the full article on the original source
Go to the source →