ABC 7 Chicago’s Caroline Foreback reports that armed officers converged on students outside Kenwood High School in Baltimore County after the campus Omnilert AI system flagged a supposed gun.
Bodycam video captures the urgency. “Keep your hands on your head,” an officer orders, weapons out and voices tight.
The teen at the center of the alert had no gun. Just a Doritos bag.
In the footage, an officer even points to the screen image: “That’s what it looks like right here.” It looked like a weapon. It wasn’t.
Police detained the group, handcuffed the student, and searched him. Nothing. No firearm. Only the crushed chips bag that had tripped the algorithm.
Officials Say the System “Worked” – Students Say Otherwise
After officers cleared the scene, school officials reviewed the alert and canceled it. Baltimore County’s Department of School Safety and Security said that was the process as designed.
Superintendent Myriam Rogers defended that setup. In her words, the program “did what it was supposed to do,” by signaling an alert for humans to review for cause.
But not everyone bought that logic.
County Councilman Julian Jones asked the question many were thinking: How do you end up with guns drawn over a bag of Doritos?
The bodycam offers its own grim coda. A responding officer, calm but blunt, sums it up. “AI is not the best.”
Podcast Panel: Smart Tech, Dumb Risks
On Gun Owners Radio, hosts Dakota Adelphia, Michael Schwartz, and Alisha Curtin walked through what likely happened.
Adelphia recounts reports that Baltimore County high schools run a network of cameras tied to an AI “gun detection” layer. If the software sees something “gun-like,” it pings both the school and police.
She notes a crucial wrinkle. She says she heard there’s a human check at the school level – someone who can glance at the image and say, “That’s not a gun.” In this case, she says a staffer allegedly cleared the image, but that message didn’t stop the police response already in motion.
That detail matters. If true, the failure wasn’t just an algorithm. It was a handoff problem – an alert that was canceled on one end and still live on the other.
Schwartz broadens it. He calls the whole thing a “comedy of errors” that happens when we treat tools like oracles and forget they misfire. He understands the instinct to respond fast at a school, but warns that both false negatives and false positives can be tragic.
Adelphia pushes the human cost. If a teen reaches into a pocket at the wrong time, thinking it’s all a misunderstanding, someone could get shot – all because a machine framed a snack as a threat. She likens it to swatting, where a false signal summons real guns and real risk.
Curtin agrees on the principle. The technology can help, but it isn’t “perfected” enough to run without human eyes – and those humans must be able to affect the response before police escalate.
Liberty Doll: The Photo No One Has Seen
Gun rights content creator Liberty Doll lays out a timeline shaped by statements from the school and police.
She identifies the student as Taki Allen and says he was waiting after football practice around 7 p.m., finished his chips, crumpled the bag, and tucked it away. Roughly twenty minutes later, eight squad cars arrived with guns drawn.
According to her summary of the school’s letter to families, administrators received an alert that someone might have a weapon. The School Safety team “quickly reviewed and canceled the alert after confirming there was no weapon.” Counselors were offered to students, and the school emphasized partnership with police.
But the police statement, as she recounts it, simply notes they responded to a report of a person with a weapon at 7:20 p.m., searched, and found no gun.
Liberty Doll points out the confusing optics: a canceled alert but a full police deployment anyway. Did cancellation fail to reach dispatch? Did someone re-trigger the call? She says Omnilert later stressed that its system only flags possible threats for human review, but the image the AI flagged has not been released publicly.
And that missing image lingers. The software allegedly saw two-handed hold with a pointed finger – but no one outside the process has seen the still frame.
In her telling, that silence says a lot.
After-Action Questions No One Should Duck
Caroline Foreback’s bodycam clips show a chain that worked fast, then backtracked. That’s better than missing a real gun. But it isn’t good when the target is a kid with a snack.
Dakota Adelphia suggests a basic fix: the photo should go to dispatch automatically. Let experienced call-takers make a visual judgment, not just run on a text alert.
Michael Schwartz adds that time and context matter. A 7 p.m. alert – with fewer adults around and kids milling after practice – might change how monitors interpret the scene. That’s exactly when false positives can feel most urgent and least challenged.
Alisha Curtin says any AI program must give humans the power to pause escalation in real time. A checkmark no one reads is not a safety valve. (Alisha Curtin.)
What Went Wrong, Technically?
From the reporting and podcasts, three failure points stand out:
Pattern confusion. The system likely latched onto the silhouette – two hands, a small object, one finger extended – and matched a known “gun pose.” That’s classic computer vision brittleness.
Alert plumbing. The school says it reviewed and canceled quickly. Yet police still arrived heavy. That implies a routing or latency bug between school consoles, district security, and dispatch.
Escalation defaults. Once the call hits the police channel as “person with a gun,” the on-scene posture is guns drawn until proven otherwise. The false positive becomes a high-risk stop, not a casual check.
Each layer makes sense by itself. Combined, they turned a trash bag into a felony stop.
Schwartz notes that searches and stops at schools operate under different standards than on the street.
That doesn’t make this harmless. It just explains how quickly the line from alert to handcuffs can be crossed in a campus environment.
Adelphia’s warning lands hardest: when the stakes are this high, the cost of a mistake is human. The right design goal isn’t “catch everything.” It’s catch real threats without creating new ones.
Safety Tech Without Trust Becomes a Threat
This story is not just about a glitchy model. It’s about trust.
Parents, students, and teachers will accept cameras and detectors when they see competence – fast, accurate alerts, clean hand-offs, and visibly smart decisions under pressure.
What they saw here was overreaction born of a misfire, then a tidy memo after the fact.
That gap breeds cynicism. Not about safety, but about the people and systems claiming to provide it.
It also shows how AI needs context. A pixelated frame of a snack bag becomes a gun because the model doesn’t understand after-practice routines, a kid’s fidgeting hands, or the difference between foil glare and a pistol slide. Humans do. But only if the workflow gives them a real interrupt button.
A Better Playbook Is Not Complicated
If schools insist on AI gun detection, here’s the minimum that Foreback’s reporting and the Gun Owners Radio discussion point toward:
Show the image to dispatch, every time. Don’t send a gun call without a human confirming the frame looks like a gun.
Lock the pipeline. When a school safety officer cancels an alert, that cancellation must preempt the police response automatically. No crossed wires.
Tier the response. If the image is ambiguous, stage nearby units, but let a school resource officer or supervisor approach first with clear commands, not immediate guns-out.
Audit releases. Share anonymized frames with parents and the public after false alarms. Transparency earns credibility.
None of this is anti-safety. It’s pro-sanity.
Caroline Foreback shows the moment a Doritos bag became a supposed gun and a teenager was cuffed for it.
Dakota Adelphia argues the fix is more human judgment up front, not less. Michael Schwartz warns that both missing a gun and creating a false gun crisis can end in tragedy. Alisha Curtin says humans need real authority inside the response loop, not just after-action notes.
Liberty Doll underscores the unresolved piece – the unseen image and the messy timeline where a canceled alert still brought out eight cars with guns drawn.
The lesson isn’t that we should ditch safety tech. It’s that tools without trust are just new ways to make old mistakes, only faster – and sometimes with kids in the crosshairs.
For more info, watch the ABC 7 Chicago clip here, the Gun Owners Radio video here, and Liberty Doll’s video here.
UP NEXT: “Heavily Armed” — See Which States Are The Most Strapped
Image Credit: Survival World
Americans have long debated the role of firearms, but one thing is sure — some states are far more armed than others. See where your state ranks in this new report on firearm ownership across the U.S.
The article Student Handcuffed After AI School Security System Mistakes a Doritos Bag for a Gun first appeared on Survival World.
Raised in a small Arizona town, Kevin grew up surrounded by rugged desert landscapes and a family of hunters. His background in competitive shooting and firearms training has made him an authority on self-defense and gun safety. A certified firearms instructor, Kevin teaches others how to properly handle and maintain their weapons, whether for hunting, home defense, or survival situations. His writing focuses on responsible gun ownership, marksmanship, and the role of firearms in personal preparedness.
