Armed police swarm student after AI mistakes bag of Doritos for a weapon (www.dexerto.com)

A Baltimore County teenager was handcuffed at gunpoint after an AI-based gun-detection system misidentified a crumpled bag of Doritos as a firearm. The October 20 incident involved multiple police cars responding to an Omnilert alert that had scanned school surveillance footage in real time; officers arrested 16-year-old Taki Allen, searched him, then released him after confirming the item was chips. Omnilert later called the event a “false positive” but said the system “functioned as intended,” pointing to rapid human verification as part of its workflow. The school district offered counseling but — according to Allen — no personal apology, and the student says he now feels unsafe returning to campus. The episode underscores core technical and policy tensions as AI surveillance spreads: systems tuned for high sensitivity will generate false positives that, when tied to armed response protocols, can produce serious real-world harms. Key implications for the AI/ML community include the need for transparent performance metrics (false-positive/false-negative rates), rigorous field testing, clearer human-in-the-loop verification standards, and deployment policies that limit escalation before visual confirmation by trained staff. Beyond accuracy, the case highlights operational design choices — alert thresholds, latency, and integration with law enforcement — that determine whether detection systems improve safety or introduce new risks.