Teen Swarmed by Cops After AI Metal Detector Flags His Doritos Bag as a Gun (gizmodo.com)

A Baltimore high-school student was handcuffed and briefly detained after an AI-guided campus security system flagged an item as a weapon — later reported to be a crumpled Doritos bag. School officials say an automated alert prompted review by the Department of School Safety, which canceled the initial alarm, but the school resource officer called local police for backup; officers searched the teen, found no weapon, and confirmed the scene was safe. Neither the school nor police have explicitly confirmed the chip bag as the trigger, but local reporting links the false positive to Omnilert, a vendor that markets AI “gun detection” for schools. The episode highlights why machine vision detectors deployed in high-stakes environments must be treated cautiously: object-detection models can confuse similar shapes, textures or reflections and suffer from domain shift (training on one set of images, failing in messy, real-world scenes), leading to false positives with serious safety and civil-liberties consequences. Key technical implications include the need for rigorous false-positive/false-negative reporting, calibrated decision thresholds, human-in-the-loop verification, independent audits, diverse training data, and transparent performance metrics under varied lighting and occlusion. Beyond engineering fixes, the incident underscores policy questions about accountability, deployment oversight, and the psychological risk to students when imperfect AI systems trigger armed responses.