Two separate cases involving teenagers in Pakistan have raised concerns about how artificial intelligence could be misused in connection with violent crime planning.
According to investigators, the two 17-year-old college students lived in Bahawalpur and Karachi and did not know each other. Both reportedly had a strong interest in American crime dramas and were allegedly influenced by fictional characters from shows such as Breaking Bad and Dexter.
In the Karachi case, investigators said the teenager had discussed an alleged plan to attack a person connected to a police officer. According to the report, he used an AI chatbot while discussing the idea and presented himself as being influenced by Dexter Morgan, the fictional forensic expert and serial killer from the television series.
The conversation was reportedly flagged by Anthropic’s Claude, which subsequently alerted the FBI. Information was then passed to Pakistani authorities, leading the National Cyber Crime Investigation Agency to investigate the teenager. The student later denied having any intention to commit murder and said his conversations were related to writing a fictional story. Investigators said no prosecution could proceed because sufficient evidence had not been established.
A separate investigation involved a teenager from Bahawalpur who was reportedly inspired by Walter White from Breaking Bad. Investigators said he developed an interest in chemistry and allegedly attempted to harm his father following a disagreement. The teenager reportedly consulted AI systems afterward, but the chatbot involved refused to provide further assistance.
The cases have prompted concerns among Pakistani investigators about the challenges of identifying potential threats in the AI era. Officials are particularly concerned about situations where fictional content, personal grievances and AI tools intersect.
The incidents also highlight differences in how AI platforms respond to requests involving violent activity. A recent industry test cited in the report found significant differences among major chatbots in their responses to dangerous requests.
