Flock says its new tool will help identify police abuse, but hasn’t explained how it works

Flock wants police departments to trust an algorithm it won't describe. The surveillance startup announced Thursday that its "Audit Assistance" tool — designed to catch officers abusing license plate reader databases — has already been activated by more than a third of its law enforcement customers. By December, every single one of them must turn it on. The company says the tool works. It says it has already caught abuse. What it will not say is how.

The contradiction sits at the center of Flock's announcement. Ashley Haber, the company's head of trust and compliance, describes the tool flagging "odd search history from a specific user." Co-founder Paige Todd offers a concrete example: a user searching the same plate repeatedly for more than 30 days. Major Patrick Krieg of the Dunwoody, Georgia Police Department goes further. He calls it "an algorithm that has notified us of any type of bias." Bias. That word carries weight. It suggests the tool doesn't just catch policy violations — it detects discriminatory patterns. That is a remarkable claim for a piece of software Flock insists contains no machine learning, no artificial intelligence, no model trained on historical data.

So what is it? A spokesperson told TechCrunch it is "a data tool that flags atypical search patterns for further review." The sentence trails off in the company's own statement: "One example that would trigger a flag is if a user searches for the same plate using multipl—" The sentence fractures there, cut off in the reporting. But the omission is telling. Flock will give examples. It will not give the logic.

This matters because license plate readers are among the most pervasive surveillance technologies in American policing. They capture every vehicle passing a fixed camera or a patrol car, building a dragnet of location history that can reconstruct a person's movements over months. The Fourth Amendment implications are unresolved. The abuse potential is documented — officers running plates for ex-partners, political rivals, personal curiosity. Flock's product sits inside that infrastructure. Its accountability layer is not a nice-to-have. It is the only technical guardrail between a database of innocent people's movements and the humans paid to access it.

And that guardrail is a black box.

Flock's blog post frames Audit Assistance as a workflow tool: surface atypical activity, review it quickly, document the outcome. Automatic lockouts for flagged users pending administrator intervention. That sounds like process. Process is not detection. The detection — the decision of what counts as "atypical" — is the whole question. Is it a threshold? Three searches of the same plate in an hour? Ten in a day? Does it weigh time of day, shift patterns, geographic clustering? Does it compare an officer against their peers? Against their own history? Does it incorporate crime data to distinguish investigative persistence from stalking? Flock will not say. The company has been asked. It has not answered.

The mandate changes the stakes. When a feature is optional, opacity is a sales problem. When it is compulsory, opacity is a governance problem. Every Flock customer — police departments across the country — must run this undisclosed logic on their own officers' queries by January. They will receive flags. They will investigate or dismiss them. They will have no independent way to verify the flagging logic works, or to explain it in court when a defense attorney asks how the department knows its evidence wasn't tainted by the very bias the tool supposedly catches.

Flock's data retention changes accompany the announcement. Seven days instead of thirty. "Evidence Mode" for exceptions, tied to case numbers. Sharing limits by offense type. These are policy knobs. They are visible, debatable, adjustable. Audit Assistance is none of those things. It is a judgment engine shipped without a manual.

The company's refusal to disclose is not a trade secret defense. Flock says the tool uses no model, no training data, no AI. If that is true, the logic is deterministic — a ruleset. Rulesets can be published. They can be audited. They can be stress-tested by independent researchers. A surveillance vendor that claims its accountability software is too sensitive to describe is asking for a trust it has not earned. The market for license plate readers is consolidated. Flock's customers cannot easily switch. They are locked into the opacity.

There is a pattern here. Surveillance companies sell certainty to police and ambiguity to the public. Flock's marketing language — "surface atypical activity early," "clear, documented workflow" — promises transparency. The product delivers a flag and a lockout. The gap between them is where abuse hides, or where false flags ruin careers, or where both happen and no one can prove either.

The Dunwoody police major called it an algorithm for bias. Flock says it is not AI. One of them is using the wrong word, or the right word for the wrong reason. Either the tool uses statistical learning the company denies, or it uses hardcoded heuristics the company fears to publish. Both possibilities demand scrutiny. Neither receives it.

By year-end, the tool will run on every Flock installation in the country. Officers will be flagged. Administrators will review. Cases will be built or dismissed. And not a single person outside Flock's engineering team will know why the flag flew. That is not accountability. That is a trust fall with no catcher.