Product1 distinct publisher3 min readPublished
Every number in this case came out of a public records request rather than an alert. Three months of queries accrued in the log before any consequence arrived, which is what buying lookup power without a detector looks like from the outside.
The Product Desk · Product desk
Compiled by The Product DeskSomething wrong?How this is made
Fifty-six searches on one plate plus 29 on another comes to 85 queries aimed at two named colleagues [5]. March through May is roughly 92 days, so the log was taking in about one search a day on people the searcher worked beside [16]. A rule that matched on one operator hitting the same plate dozens of times in a quarter would have had plenty to match on.
WIRED's account describes no point at which anything in the tooling interrupted the searches [19]. The first documented consequence is paid leave. Even on the most generous reading, where the final query landed on May 31, that is 50 days, about seven weeks, before the suspension on July 20 [15]. If the last search came earlier in May, the gap is longer.
Here is what teams tell themselves the search box is for: cases. Here is what people do with a search box: they type in the question they actually have. Bozzo told the internal investigator he had not used Flock with intent to intimidate or cause harm [11]. Whatever you make of that, it states the product problem precisely. At query time the console cannot tell a case from a curiosity, and both increment the same counter. On a usage dashboard, 85 searches from one patrolman is a well-adopted tool.
The detection work in this story was done by people. Colleagues had already reported the pair spending personal time together on duty while not doing their assigned work, which got them separated from the same squad and beat and handed a written directive about acceptable workplace interactions [12]. A sergeant reported seeing them meet in a dark, secluded park at about 2:30 am while Bozzo was on duty [13]. That is anomaly detection performed with eyes and memos, and the national tally WIRED keeps is in incidents rather than queries [10], because an incident is the unit that surfaces when a human files something.
Two questions make the grid worth drawing for your own systems. Does the system record the lookup, and does anything read that record when nobody has complained. Most organisations sit in recorded-and-unread, which is a forensic capability sold as a safeguard: excellent at producing an exact number for the report, useless while the number is climbing.
The version you can run tomorrow is narrower. List the identities your own staff would most want to look up, and check whether a lookup on one of them is distinguishable, at the console, from a lookup on a stranger. Then name the last alert the system generated and who acted on it. Where no alert exists, the control is a colleague willing to write a report about someone they work with, which is what Alpharetta had, and it took from March to July [2][6].
Ranked by verification strength, evidence, and original report placement.
Across the country there have been at least 45 documented incidents of police officers misusing Flock to look up or track current or former romantic partners.
According to the documents, which include the full internal report written by Alpharetta police Sergeant Jabbar Braithwaite, Bozzo says he did not use Flock with the intent to intimidate or cause harm.
WIRED obtained documents via a public records request revealing specific details of a Flock-based surveillance campaign by a police officer, plus the subsequent investigation and fallout.
Dustin Bozzo, who until this month worked as a patrolman in Alpharetta, Georgia, allegedly used Flock 56 times between March and May to search for the license plate of a former romantic partner who was a fellow member of the Alpharetta Police Department.
The relationship between Bozzo and the other officer itself violated the city's ban on workplace dating, according to the documents.
When Bozzo noticed that his ex's car was often near that of another police officer, he began searching for that officer's vehicle as well, doing so 29 times between March and May.
Distinct publishers with included, body-backed reporting in this cluster.
1 article · August 27, 2026
Follow any of these and your For You feed starts watching them — no settings page required.
product
LAPD's Flock reset shows the contract, not the camera, is now the product1 distinct publisher
security
Audit logs caught a Florida deputy's Flock searches only after his ex-girlfriend complained1 distinct publisher
build
Flock's OS Investigate starts the search before there is a suspect1 distinct publisher
security
A man in a Vader helmet turned Flock's San Diego contract into a procurement problem1 distinct publisher
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Primary documents and named officials, but one outlet and two silent parties
The core factual spine rests on released public records including the full internal investigation report and Flock audit-log citations, with two named Alpharetta officials speaking on the record and specific dated actions. Weakening it: a single publisher, no independent verification of the query counts, the subject and the vendor both declined comment, and the nationwide 45-incident figure is a count of surfaced cases with no denominator.
Live agency deployment with logged per-query usage, plus a multi-case national pattern
Adoption is directly observable: Flock is in production use at Alpharetta PD with individual officers holding instant query access, per-query audit logs and reason prompts running in the deployed product, and at least 85 queries recorded by one user in three months. The nationwide floor of 45 documented partner-tracking misuse incidents implies broad multi-agency deployment. The score is held below high because the source gives no agency count, contract detail, or total query volume.
Slightly understated: documented counts are floors and the vendor side is unanswered
The reporting is conservative relative to what its own documents show. The headline count of 85 excludes 18 additional searches of a relative's vehicle, the national 45-incident figure is presented as a documented minimum in a field where 'details are often scarce,' and the story stops at one agency rather than asserting a systemic detection failure. Nothing in the material overstates capability or consequence; the only inflationary language is the department's own 'immediate action' framing, which sits against a roughly seven-week gap between the last documented search and the first personnel action.
Strong self-interest on every named side; vendor incentive unexpressed
Each first-hand account carries visible motive. The department, facing an open criminal investigation, has direct reputational reason to stress its 'high standard' and 'immediate action.' The officer, subject of that investigation, denies intimidating intent, attributes his reason-code choices to convenience, and then altered his social profile after being contacted. Flock's incentive is unexpressed because it did not respond. WIRED's own accountability beat, signalled by the framing of a surge in misuse reports, favours cases that fit that pattern.
High confidence in the documented timeline, lower in systemic scope
The case-level facts - query counts, dates of leave and resignation, the workplace-dating violation, the open criminal investigation - are documentary and attributed, so confidence in them is high. Confidence drops for the wider reading: single-publisher sourcing, no vendor response on detection capability, no denominator for query volume or agency count, and a national figure that is an acknowledged floor. The derived no-detector conclusion is an inference from absence in one published account rather than a verified control audit.