# Maryland students sue over AI gun alert on Doritos bag

> Four Maryland students sued Omnilert, police and school officials after an AI gun-detection alert mistook a Doritos bag for a firearm. The Guardian reports the claims; the defendants have not answered them.

Published 2026-11-19, updated 2026-11-19. Confidence: medium. When it could matter: Timing unknown; depends on the court process and the defendants' responses.

## In plain words
Four current and former students at a Maryland high school have sued after an AI camera system flagged a bag of Doritos as a gun. Armed police then detained and handcuffed them while they waited for rides home. The students say the technology often gives false alarms and that adults failed to stop the response once the mistake was clear. The claims are allegations, not findings. The case shows why an AI alert should be checked by a person before armed officers act on it.

## What happened
The Guardian reports that the four plaintiffs sued Omnilert, the maker of the AI gun-detection system, along with Baltimore county police, the board of education, individual officers and the school principal. They allege unlawful detention and search resulting from technological and human errors.

According to the complaint, the incident took place on the evening of 20 October last year. The students had finished football practice and their coach had handed out bags of Doritos. Taki Allen, then a junior, was near the picnic tables with friends waiting for a ride when six police cruisers arrived with lights flashing. Many officers unholstered their weapons, and the students put their hands in the air.

The complaint says Omnilert's Gun Detect system, which scans security footage for signs of firearms, flagged the chips bag as a firearm with 90% confidence. It says the alert continued after school staff and local police reviewed the image and decided it was a mistake. The lawsuit also cites Maryland's office of the inspector general for education as finding that more than 99% of alerts from the system in Baltimore County Public Schools were false positives needing no police response.

The Lawyers' Committee for Civil Rights Under Law represents the students. Its president, Damon Hewitt, said the students were failed by a technology prone to false alerts and by the adults meant to keep them safe.

## What is confirmed and what is not
**Confirmed by the source:** a lawsuit has been filed, naming the defendants listed above. Omnilert told the Baltimore Banner that it was reviewing the suit and was sorry for the fear the families experienced. A Baltimore county spokesperson told the New York Times the district cannot comment on pending litigation.

**Claimed, not established:** the sequence of events, the 90% confidence figure, the statement that the alert continued after review, the officers' remarks, the principal's messages and the inspector general's 99% finding are all taken from the complaint as reported. The Guardian does not say it has seen the inspector general's report itself. The students' reported pain from handcuffs and lasting distress are their own account.

**Unknown:** the defendants' legal responses, the specific legal grounds and remedies sought (the Guardian's account does not set them out), and how the alert passed between the software, school staff and police. Omnilert, the principal, the police and the board of education did not immediately respond to the Guardian.

## Analysis
### Armed police acting on an alert that people had already judged wrong
The most direct consequence is physical risk to people who did nothing wrong. According to the complaint, the system produced a false alert, and the harm followed when the alert kept going after school staff and police had reviewed the image and decided it was a mistake. If that is accurate, the central failure is not only the detection error but the missing step where a human judgment overrides the machine and stops the response. The complaint also reports an officer saying thanks for not running away, which, if true, hints at how close the situation was to something worse. Hewitt said it could have been even worse.

This is TOSWO Research's reading of the allegations, not an established finding. A sceptical expert would ask who held authority to cancel the alert, how a cancellation travelled to the officers, and whether the officers were told the image had been reviewed. The Guardian's report does not answer these.

### Law and policy
The plaintiffs allege unlawful detention and search. In general terms, a claim like this requires showing that officers stopped and searched people without adequate legal justification. The justification here would depend on what the officers knew and when. The source does not specify the legal provisions invoked or the remedies sought, so those cannot be assessed. The defendants are not one block. Omnilert is sued as the maker of the system, while the police, board, officers and principal are sued for decisions that people made. The plaintiffs frame the harm as both technological and human error, and the complaint's weight may differ between them.

The link between the AI error and the harm is shown by the complaint's account, but only asserted until tested. A court would likely look at whether the company's marketing, which the lawsuit says promises "fewer false positives" [1], matches the reported performance. The inspector general figure matters here, because a false positive rate above 99% would mean that most alerts needed no police response. If that figure is accurate, then a response protocol that sends armed officers by default would be built on a signal that is mostly wrong. If it is wrong or measured differently, that argument weakens.

### Technology and industry
The numbers illustrate a general problem with detection of rare events. When real weapons are very rare, even a system with high stated confidence on a single image can produce mostly false alerts overall. That is a general principle, not a finding of the source. The 90% confidence number alone does not tell a reader how often such a score corresponds to a real gun. For schools and vendors, the useful questions are the measured false positive rate in deployment, what a human reviewer sees, and who is allowed to dismiss an alert.

The source gives no information on whether Omnilert's system includes a human review step on its side, and the lawsuit's account that the alert kept going after local review implies either no way to cancel it or a failure to use one. That is a reading of one side's complaint, and Omnilert's reply could change it.

### Daily life
The students report lasting effects. Taki Allen says he no longer feels safe waiting for rides or staying at school after hours and often pays for rides home. For families, the practical lesson is that AI-based security tools in schools can carry real consequences for children, and that parents may want to ask how alerts are verified. Similar failures involving automated systems acting without human checks appear in the wider record of AI incidents, such as the [Anthropic test agent that filed a false tip with police](/events/press-anthropic-haiku-false-tip-2026-10), although that case involved a different kind of system and setting.

### Limits of this reading
The strongest reason this view could be wrong is that it rests on a complaint, which is one side's account written to persuade. The defendants may dispute the timeline, the 90% figure, or the inspector general's finding, and a review could show that officers acted on information the complaint omits. Evidence that would change the reading includes the police dispatch records and any evidence of when the alert was cancelled. This story rests on a single report, so each claim here is limited to what The Guardian reports.

### What would settle it
The complaint itself, with its exhibits including the principal's texts, would show exactly what is alleged and on what legal grounds. The inspector general's report would confirm or correct the 99% figure and its definition of a false positive. The defendants' formal responses in court, and any Omnilert technical documentation on alert handling and cancellation, would show where the process failed. Police records of the dispatch would show what officers were told.

## What can be done
For schools and districts using AI detection: require a trained person to verify each alert before police are sent armed, publish the false positive rate, and set a clear way to cancel an alert that reaches police. For police: confirm with the school whether an alert has been reviewed before approaching. For parents and students: ask the school how alerts are handled. If stopped by police, staying calm and following instructions is the general safe course, though this is not legal advice. Readers who want to follow wider patterns can look at the [incident record](/record) and how TOSWO [measures AI risk](/method).

## Questions readers ask
### What did the AI system do in the Maryland Doritos case?
According to the complaint, Omnilert's Gun Detect system flagged a chips bag as a firearm with 90% confidence. The lawsuit says the alert continued even after school staff and police reviewed the image and decided it was a mistake.

### Who is being sued?
The plaintiffs sued Omnilert, Baltimore county police, the board of education, individual officers and the school principal. They allege unlawful detention and search because of technological and human errors.

### How often are these alerts wrong?
The lawsuit says Maryland's office of the inspector general for education found that more than 99% of alerts in Baltimore County Public Schools were false positives that required no police response. This is as cited in the complaint, and the Guardian does not report a response to it.

### Has Omnilert responded?
Omnilert told the Baltimore Banner that it was reviewing the lawsuit and was sorry for the fear the families experienced. It did not immediately respond to the Guardian's request for comment.

## Who is affected

- Students waiting outside schools where AI gun detection is used (relative exposure 9 of 10)
- School districts and boards deploying AI weapon detection (relative exposure 8 of 10)
- Police officers responding to automated school alerts (relative exposure 7 of 10)
- AI security vendors like Omnilert (relative exposure 6 of 10)
- Parents of schoolchildren (relative exposure 5 of 10)

## Risks

- Armed officers respond to false AI alerts and endanger innocent students
- Alerts keep running after humans have judged them mistaken
- Very high false positive rates make real alerts harder to trust
- Children avoid staying at school after hours, with lasting distress

## What can be done

- Require human verification of each alert before armed police are sent
- Give schools a clear way to cancel an alert that has reached police
- Publish measured false positive rates for deployed systems
- Train officers to confirm whether an alert has been reviewed before approaching

## Sources

1. Four Maryland high school students sue after AI mistakes Doritos bag for gun, The Guardian, 2026-11-18: https://www.theguardian.com/us-news/2026/nov/18/maryland-baltimore-county-doritos-gun-ai-lawsuit

---
Source: https://www.toswo.com/signals/maryland-students-sue-over-ai-gun-alert-on-doritos-bag
Prepared by Toswo Research (https://www.toswo.com/research) from the sources above. Free to cite with attribution: https://www.toswo.com/cite
