Doritos Bag Flagged as a Gun: Teen Cuffed by Police

Illustration of a student holding a blue snack bag at a school entrance at night, framed by a red AI warning scan box, with police car lights flashing behind

They Were Just Eating Doritos, and Police Pointed Guns at Them

Around 6:30 p.m. on October 20, 2025, it was calm outside Kenwood High School in Baltimore County, Maryland.

Taki Allen, a 16-year-old who'd just finished football practice, was waiting for a ride home and chatting with three teammates: Yui-Osa Ogbeide, Devon Thompson Jr. and Keith Winfield III.

Taki was holding a bag of Cool Ranch Doritos that the coach had given them after practice, and they were laughing, talking and sharing the snack.

Then at 7:21 p.m., police were ordered to respond, and around 7:23 p.m. officers arrived. The scene in front of the school turned into a horror movie in seconds.

Several Baltimore County police cars rushed into the school's front lot without warning, and officers with their guns drawn shouted at the students as they closed in.

Taki and the three friends put their hands up with no idea what was happening. Police made them kneel on the concrete and handcuffed them behind their backs.

In a media interview, Taki said, "When I saw the police pointing guns at me, the first thing I thought was, am I going to die?" Four teenagers were held in handcuffs on the cold ground for at least 6 minutes and searched.

But even though police searched the students themselves, they found nothing. An officer then checked a trash can near where the students had been and found several empty Doritos bags.

〈WBAL-TV report on the Doritos bag gun scare〉
* From the official WBAL-TV 11 Baltimore channel.

The Absurd Story of an Alert Canceled in 2 Minutes and Revived 10 Minutes Later

✦ Timeline of the Day of the Incident

The evening of October 20, 2025

Around 7:04 p.m.
AI flags a suspected gun
7:05-7:06 p.m.
Safety director and police check and cancel the alert
Around 7:19 p.m.
Principal texts district executives
7:21 p.m.
Police ordered to respond
7:23 p.m.
Police arrive at the scene

The reason armed police showed up in front of unarmed teens was Omnilert Gun Detect, an AI gun detection system hooked up to the school's security cameras.

At around 7:04 p.m. that evening, the system spotted Taki holding a snack bag near the school entrance and raised a "suspected gun alert."

✦ What Did the AI Detect?

The OIGE investigation report and the complaint tell different stories

OIGE Investigation Report
• Didn't detect 'a single blue bag of chips' as the threat
• The position and angle of a hand gripping an unidentified object sticking out past the fingers triggered the alert
The Students' Complaint
• Claims the system identified the Doritos bag in Taki's hand as a '90% probability gun'

There's still a debate over what the AI actually detected. The Maryland Office of the Inspector General for Education (OIGE) investigation report stated clearly that the AI didn't simply flag "a single blue bag of chips" as a threat.

What triggered the alert was the position and angle of a hand gripping an "unidentified object" that stuck out slightly past the fingers, at about waist or pocket height.

But the students' complaint filed in court claims the system identified the Doritos bag in Taki's hand as a "gun with 90% probability."

Just because the AI raised an alert doesn't mean police automatically roll out. Omnilert's system is designed so that a human review team inside the company checks the screen first.

But the human reviewer didn't catch that the object on screen was a snack bag either, and approved it as a real threat. There's one more step, though. The alert goes out to district staff and police, and it can get filtered there too.

In fact, the alert text went to at least 15 phones: 5 Baltimore County Public Schools (BCPS) staff and 10 county police department officials. The text included a link to a web page with a photo of Taki holding the snack bag and a short video clip.

But between 7:05 and 7:06 p.m., BCPS Executive Director of School Safety Eric Knox, BCPS safety officer Marvin-Basta and police Sgt. Robinson watched the video and realized it was "a snack bag, not a gun."

So they canceled the alert in just 2 minutes. The linked web page switched to "canceled," and a cancellation text went to everyone who'd received the first one.

What should've ended there came back to life more than 10 minutes later, from an unexpected place. Principal Catherine "Kate" Smith of Kenwood High School, one of the first alert's recipients, finally opened that first text.

The principal missed the cancellation text that arrived right after, and didn't notice the cancellation mark on the enlarged video screen either.

Next, the principal tried calling a district school safety officer to check the alert's status, but nobody picked up.

So the principal contacted School Resource Officer (SRO) John Douglas on a personal cell phone, asked about the alert and sent a screen capture of Taki holding the snack bag.

So around 7:19 p.m., Principal Smith texted district executives: "Omni alert went off, and an apparently armed student of ours is standing on the school terrace, so the SRO is reporting it to the precinct."

When the SRO passed the screen capture on to police, an alert that had already been canceled in the official system more than 10 minutes earlier turned into a brand-new report.

It ended up bringing at least 6 armed patrol cars (about 8, going by the students who were there).

(Ahh... Principal, why did you do that...)

〈NBC report on Omnilert AI gun detection〉
* From the official Omnilert channel.

7,000 Cameras and a Police Notification Rate Under 1%

✦ Omnilert's Rollout Scale and Cost

Baltimore County Public Schools

CamerasAt least 7,000
Contract term3 years 9 months, with a separate 1-year extension option
Spending authorization cap$2,663,000
Annual monitoring cost$560,000

Starting in fall 2023 and running through spring 2024, Baltimore County Public Schools hooked Omnilert's system up to at least 7,000 existing security cameras across its 165 schools.

The 3-year, 9-month contract, which includes a separate 1-year extension option, has a spending authorization cap of about $2.6 million (the official cap is $2,663,000), and separate budget documents set aside $560,000 a year for monitoring.

Omnilert doesn't use facial recognition. It analyzes camera frames in real time to catch guns and postures close to gun violence, and it's been promoting near-zero false alarms by combining AI with human verification.

Right after the incident, Taki's family and the local community raised suspicions of algorithmic discrimination targeting Black students. The OIGE launched an official investigation and released its report on November 25, 2025.

The inspector general's office found that the AI algorithm weighs only moving objects and hand posture rather than factoring race or gender into its probability calculations, so it concluded that evidence of racial bias in the software itself was hard to find.

✦ Alert-Handling Statistics

Recorded from November 2023 to November 2025

Stage 1, of the objects the AI detected, those filtered out at the monitoring stage with no further action needed
96%
Stage 2, of the alerts passed on to the school district, those where notifying police was needed
Under 1%
Stage 2, of the alerts passed on to the school district, those where notifying police wasn't needed
Over 99%

The alert-handling statistics the inspector general's office released alongside the report were eye-catching too.

Looking at two years of records from November 2023 to November 2025, 96% of objects the AI detected in the first stage were filtered out at Omnilert's monitoring stage and needed no further action.

Looking only at the second stage, alerts that passed Omnilert's verification and went to the school district, police notification was needed in under 1% of them.

In other words, over 99% of the alerts passed on to the district were cases where notifying police wasn't needed.

Each figure is based on a different processing stage and denominator, and alerts that didn't need police notification and false alarms are metrics that have to be read separately.

✦ Kenwood High School vs. Antioch High School

The same Omnilert system

Kenwood High School Incident
• An alert went off over a snack bag
• It led to armed police being dispatched
Antioch High School, January 2025
• Nashville's school district spent over $1 million to install the same system in its schools
• The shooter was far from the camera and the gun was concealed, so it went undetected beforehand
• Casualties followed

This system's weak spot had already been exposed back in January 2025. Nashville's school district in Tennessee spent over $1 million to install the same Omnilert system in its schools, and Antioch High School was one of them.

But because the shooter was far from the camera and the gun was concealed, the system failed to catch an actual shooting beforehand, and there were casualties.

It threw up an alert that brought armed police over a snack bag, but it couldn't catch a criminal coming in with the weapon hidden.

The Developer Says the System "Functioned as Intended," and the Students Take It to Court

Right after the incident, Omnilert spokesperson Blake Mitchell said, "The nighttime lighting, the bag's color and glare, and a posture with one finger extended looked a lot like holding a gun, so our human verification team confirmed it and passed it on."

The company then defended the whole process as having functioned as intended.

The school and police did a joint review and promised annual retraining, but the officers who responded to the scene and handcuffed innocent students got away without any discipline.

✦ Lawsuit Overview

Filing date
November 17, 2026
Court and complaint
Circuit Court for Baltimore County, Maryland; 31 pages
Plaintiffs
4 students at the time, including Taki Allen
Defendants
Omnilert, the school district, Principal Smith, the responding police officers
Relief sought
Damages and an injunction to prevent recurrence

So about one year and one month after the incident, on November 17, 2026, four students from that night, including Taki Allen, began a legal fight.

Taki filed through guardians Jerrie Davis and Lamont Davis, and the students got help from groups including the civil rights organization Lawyers' Committee for Civil Rights Under Law.

The 31-page complaint (Allen, et al. v. Omnilert, et al.) filed in the Circuit Court for Baltimore County, Maryland, names Omnilert, the school district, Principal Smith and the responding police officers as defendants side by side.

The plaintiffs argue that police restrained and searched minors without even a reasonable basis for suspicion, violating the Fourth and Fourteenth Amendments to the United States Constitution, the Maryland Declaration of Rights and the prohibition on false imprisonment.

The plaintiffs also pointed out that "most alerts passed on to the school district didn't need a police response" and that "letting the cancellation mix-up slide put students in danger," and demanded damages and an injunction to prevent it from happening again.

How Far Can We Trust AI?

Lately, with AI technology advancing, some people make the extreme argument that "judges and the judiciary should be replaced with AI." Their point is that judges are human too, and they often rule based on their own beliefs or emotions.

AI, they say, would judge objectively and fairly.

But look at this Doritos bag incident, and you can see AI makes critical mistakes just like humans do. Human judges also turn innocent people into criminals with a split-second misjudgment or a ruling that isn't neutral.

But AI is the same. AI runs on algorithms, and no algorithm is perfect, right? So AI makes mistakes in some cases too.

For example, AI is being brought into airport security screening too, but there's a catch. If you turn up the AI's sensitivity for catching dangerous items, it flags way too many things as dangerous, and if you turn it down, it misses dangerous items.

The gun detection system in this case is probably the same. Turn up the AI's sensitivity for finding guns and it classifies absurd objects as guns, like it did here. Turn it down and it misses guns, which leads to accidents.

In the end, there's still a lot lacking in handing important judgments to AI alone. Humans have to act as the verifiers, but in this case the humans didn't do that job properly.

That's all for this post. See you next time.

AI-Faked Photo Won a DoorDash Pizza Refund. CCTV Caught It

ChatGPT Lawsuit: "Cut Your Dose, Hide It From Your Doctor"

AI Scam Gang Posed as CEOs and Stole $129 Million

AI-Faked Photo Won a DoorDash Pizza Refund. CCTV Caught It

A Pizza Burned Black and an Automatic Refund On the morning of October 23, 2026, Rhonda Reynolds, who runs a pizzeria in Echo Park, Los Ange...

Removing Samsung TV Ads With AI Got a Codex Account Blocked?

The Person Who Got Suspended From Codex While Poking at a Smart TV You turn on the smart TV sitting in the middle of your living room, and a...

ChatGPT Lawsuit: "Cut Your Dose, Hide It From Your Doctor"

A Valentine's Day Night, and a Tragedy Seen Over a WeChat Video Call ✦ Key Timeline of the Case November 2025 to November 2026 Nov 13, 2...