Nathaniel Michael Carrasco, 22, was arrested by San Antonio police after Anthropic’s AI chatbot Claude flagged his alleged threats to attack a neighboring elementary school and alerted federal authorities.
The FBI’s National Threat Operations Section notified San Antonio police on Aug. 28 about threatening queries made through Anthropic’s platform on Aug. 11, according to an arrest affidavit. Companies like Anthropic maintain strict internal guardrails designed to detect violent threats, child exploitation, and terrorism, routing credible threats to the FBI’s National Threat Operations Center or regional fusion centers.
Among the disturbing searches, Carrasco allegedly requested that the bot find images of “dead school kids bleeding,” the affidavit states.
He allegedly grew more explicit as the conversation continued, declaring that he was “getting a gun as soon as I can” and planned to walk into Serna Elementary and begin “shooting up kids,” according to KENS5. “I truly do mean my threats,” he allegedly wrote in another message.
The alleged target sat alarmingly close to home. Carrasco lives next to Serna Elementary School, making the reported threats both specific and geographically immediate.
The FBI initially made a tentative identification of Carrasco as the person behind the messages. The Southwest Texas Fusion Center, which coordinates information sharing and public safety analysis, then secured emergency disclosures from T-Mobile, Charter Communications, and Google that authorities said positively identified him.
San Antonio police arrested Carrasco on Aug. 29 shortly after 11 p.m. He faces a felony charge of making a terroristic threat and is being held in Bexar County jail on $250,000 bond. If convicted, he could face between two and 10 years in prison.
The arrest, however, exposed a troubling gap in communication with the school itself. Serna Elementary School principal Jennifer Lomas addressed families on Aug. 31 after details began surfacing in news reports, telling them she wanted parents to hear about the situation “direclty from me,” according to KENS5.
Lomas said the school had received only partial information and did not learn that the specific alleged threat was directed at Serna until the details became public through media coverage.
San Antonio police said detectives began investigating immediately upon learning of the credible threat on Aug. 28, obtained a warrant, and arrested Carrasco the following night.
The incident highlights how AI safety protocols are increasingly becoming a frontline defense in identifying potential mass violence, with automated systems now serving as intermediaries between threatening individuals and law enforcement. Anthropic’s detection of Carrasco’s alleged queries represents a growing reliance on technology companies to monitor and report dangerous content generated on their platforms.
Carrasco remains in custody as the case proceeds through the Bexar County criminal justice system.