How an AI chatbot alerted FBI to two alleged murder plots in Pakistan

Two 17-year-old college students in Karachi and Bahawalpur have been investigated over separate alleged murder plans in which they reportedly discussed their intentions with an artificial intelligence chatbot, according to investigators.

The teenagers, who live about 820 kilometres apart, did not know each other, investigators said. Both were reportedly interested in American crime dramas, with investigators drawing links between their alleged plans and characters from Dexter and Breaking Bad.

According to investigators, the Karachi teenager allegedly discussed a plan to kill the son of a police officer whom he claimed had insulted his father. The teenager reportedly told an AI chatbot that he was deeply affected by the alleged incident and discussed how the fictional character Dexter Morgan might have dealt with the situation.

The teenager was later traced by the National Cyber Crime Investigation Agency (NCCIA) after the information was shared with Pakistani authorities following a report to the Federal Bureau of Investigation (FBI) by Anthropic, the company behind the Claude chatbot.

During the investigation, the teenager denied planning a murder and told authorities that he had been discussing the subject while working on a story. His father also denied that his son was involved in criminal activity.

According to investigators, the teenager was a college student pursuing an IT diploma and spent much of his time at home watching television dramas.

NCCIA officials said no criminal case could be pursued against him because the investigation had not established sufficient evidence of an actual offence. However, investigators remained concerned about the content of the conversations and continued monitoring the case, according to the report.

In the separate Bahawalpur case, investigators alleged that another 17-year-old had developed an interest in chemistry after watching Breaking Bad and subsequently attempted to poison his father following a disagreement.

The teenager allegedly established a small laboratory at home and obtained chemicals for experimentation. Investigators said he later contacted Claude about the incident after his attempt failed.

When the chatbot refused to provide further assistance, the teenager reportedly told it that he would seek information from other AI platforms. According to the investigation, he subsequently consulted other chatbots, including Grok and ChatGPT.

Anthropic reported the conversation to the FBI, which subsequently shared the information with Pakistani authorities, according to investigators.

The Bahawalpur teenager also denied planning to kill his father when questioned by authorities. His father supported him during the investigation, and a court later granted the teenager bail after his father forgave him, according to the report.

The cases have raised concerns among investigators about the potential use of AI tools by young people to explore violent scenarios.

An industry-wide study cited in the report found significant differences among AI platforms in their responses to requests involving violent-crime planning. The study assessed 10 platforms and reported that several provided assistance in such scenarios, while others generally refused.

The report also cited other international cases involving the use of AI chatbots in connection with alleged plans for violence.

The NCCIA has expressed concern over the difficulty of anticipating how individuals could use AI tools in future criminal activity. An agency official said authorities needed to consider potential threats before they materialised, according to the report.