Taki Allen’s Doritos bag shows how school AI surveillance becomes a police risk

In the article republished by Techdirt from The Conversation, the most important detail is not a technical specification. It is an ordinary object in a student’s pocket. In Baltimore County, Maryland, on Oct. 20, 2025, 17-year-old Taki Allen was sitting outside his high school after football practice when an AI-enhanced surveillance camera allegedly flagged him as a threat. The system mistook a Doritos bag in his pocket for a gun.

That kind of mistake is often called a false positive in vendor language and policy memos. In real life, the phrase is too tidy. A false positive in a security system does not stay inside a spreadsheet; it sends an alert, activates a protocol, focuses police attention and can create a record that follows a person after the danger has been disproved. For a student sitting outside school, the difference between a snack bag and a weapon is not abstract. It shapes how armed authority approaches a child.

That is why the narrow question, “How accurate was the camera?” is not enough. The harder question is who checks the machine’s conclusion before it becomes a police response. If a system is persuasive enough to accelerate intervention but too opaque to explain why it failed afterward, the public gets a dangerous bargain. Automation does not replace human judgment; it pushes it toward the most severe available interpretation.

The case matters because it sits at the intersection of school surveillance, public safety and criminal justice. AI surveillance in that setting is not looking at clean laboratory inputs. It is looking at teenagers, bags, pockets, sports gear, snacks, shadows and bad camera angles. Any of those can become a “signal” if the system sees a pattern where none exists. Once that signal enters a police workflow, the error acquires the weight of the state.

So this is not simply another story about imperfect computer vision. In the broader context of the NIST AI Risk Management Framework, systems like this demand documented risk assessment, clear accountability and procedures for challenging bad outputs. Without those layers, affected people may only learn that AI was involved after they are already dealing with the consequences.

The article was published under a Creative Commons license, which fits a story that deserves to travel beyond a narrow technology audience. The debate over AI surveillance is not whether schools may use cameras or whether police may respond to threats. The debate is whether an unverified machine signal should get privileged status in a chain of decisions that can lead to arrest, prosecution or wrongful conviction.

The most dangerous part of these systems is not cinematic autonomy. It is routine compliance. A camera flags something, software ranks it, an operator trusts it, police respond. If the “weapon” turns out to be a bag of chips, institutions cannot treat that as a minor technical mishap. It is a test of the whole system: can it admit that automated suspicion was wrong before it changes someone’s life?