Police handcuffed a high school student in Baltimore County after an AI-powered security system mistook his Doritos bag for a gun.
The incident occurred on October 20, 2025, at Kenwood High School in Maryland, when 17-year-old student Taki Allen was waiting for a ride home after football practice. The AI surveillance system, developed by Omnilert, detected what it believed to be a firearm and automatically alerted both school staff and local law enforcement.
Police officers quickly arrived at the school with guns drawn, ordering Allen to get on the ground. Allen said he was told to kneel, place his hands behind his back, and was handcuffed as officers searched him. Instead of a weapon, they found only a Doritos bag. “It was terrifying,” Allen said, describing how eight police cars surrounded the school as officers pointed their weapons toward him.
Allen later learned that the system had flagged his posture, holding the snack bag with both hands and one finger extended, as resembling someone holding a firearm. The image that triggered the alert was later shown to him, confirming the AI’s mistake.
School safety officials quickly reviewed the surveillance footage and confirmed that no weapon was present, but the heavy police response had already taken place. Baltimore County Police and Kenwood High School administration later apologized to Allen and other students who witnessed the frightening scene.
Principal Kate Smith sent a letter to parents explaining the situation and assuring them that counselors would be available to help any students affected by the event.
Omnilert, the company behind the AI detection software, expressed regret over the incident and said it had reached out to Allen and the school community. The company maintained that the system performed as programmed. However, the incident started the debates over the reliability and ethical oversight of AI surveillance in schools.
It has also brought criticism from parents and privacy advocates, who argue that while AI-based safety systems aim to prevent violence, they also risk creating new dangers when machines misinterpret everyday behaviour.
