ChatGPT flags Argentine teen's school attack plot, leading to FBI-assisted raid

Argentine Federal Police raided the home of a 15-year-old in Quilmes, Buenos Aires Province, on Thursday after OpenAI's safety monitoring flagged conversations in which the teenager asked ChatGPT what he would need to carry out an attack on his classmates, planned for 2027. The FBI's legal attaché at the U.S. Embassy in Argentina relayed the alert to local authorities after OpenAI's systems surfaced the exchange, according to Alejandro de Mena, a legal assistant at the Quilmes Juvenile Criminal Prosecution Office, who described the case to Radio Mitre.

How the alert reached police

The investigation began on August 19, when the FBI's Buenos Aires legal attaché passed along a tip generated after the teenager used monitored keywords in his exchanges with the chatbot. Investigators traced the account to the teen through his IP address. Two days after the alert, officers raided his home and seized two cell phones, tactical gloves, a military-camouflage hat, two balaclavas, and tactical protective goggles — no firearms were found. The teenager was charged with "public intimidation" and summoned to the prosecutor's office; neither he nor his parents have given testimony about the incident.

Part of a wider protocol

De Mena said the case followed a monitoring protocol the Quilmes prosecutor's office put in place earlier this year, after a wave of public-intimidation cases tied to viral internet challenges "exploded" in March and April, leading to a string of raids and prosecutions in cases prosecutors judged to be genuinely punishable rather than idle talk.

Why this case is different from other AI safety stories

Most AI safety coverage this year has centered on systems misbehaving — models breaking out of test environments, breaching platforms, or being accused of unsafe behavior by their own researchers. This case is the inverse: an AI monitoring system worked as designed, flagged a specific, escalating threat before it materialized, and triggered a cross-border law enforcement response in a matter of days. OpenAI has faced separate lawsuits and criticism this year over cases where its models allegedly failed to intervene in mental health crises; this incident is a rare instance where the company's account of its own safety tooling is corroborated by an independent government agency's timeline.

It also illustrates the tradeoffs of that monitoring: the system relies on keyword-based flags in chatbot conversations that are, by design, private, raising the same privacy-versus-safety tension that has followed content moderation on every major platform — a tension that is only sharper when the platform is answering a user's private, unfiltered questions.

As first reported by the Buenos Aires Herald.