Flock’s Camera Count Was Roughly Right. The Rest of the Network Is the Story.

A security researcher’s map of Flock Safety’s nationwide network is the subject of a takedown demand, and the number attached to it in most coverage is wrong. The underlying dataset is published openly. Read directly, it does not show a company operating twice as many cameras as it admitted. It shows something less sensational and considerably more useful: a camera count that roughly matches what Flock has said publicly, sitting inside a device network that is not made of cameras.

What the file actually contains

Cybersecurity researcher Joshua Michael reported finding an access token on Flock’s servers in November 2025 that required no login, allowing queries against the mapping platform holding device records. He says he emailed Flock on November 13, 2025, followed up twice, received one reply saying the company was internally triaging, and heard nothing further. He downloaded the data in December 2025 and published a technical write-up in January 2026.

The published dataset holds 335,701 rows. That figure has been reported as the size of the exposure, which is accurate. It has also been loosely repeated as a count of cameras, which is not. Every row carries a status field, and the rows divide three ways: 214,073 in service, 83,427 in planning, and 38,190 decommissioned, with 11 unknown. Roughly a third of the file describes hardware that was either never deployed or has already been taken down.

The camera count holds up

Filtering to Flock’s camera models, the Falcon family plus its Highway and Flex variants, and counting only devices listed as in service, the total is 123,638.

In the summer of 2026, Flock told reporters it operated more than 120,000 cameras nationwide. Measured against the company’s own device records, that statement appears to have been broadly accurate.

The widely circulated claim that the network is roughly twice the size Flock admitted comes from counting all 183,858 Falcon-family rows, including 41,281 devices marked as planned and 18,939 marked as decommissioned, and in some retellings from folding in non-camera hardware as well. Stating that plainly is not a defense of the company. A surveillance vendor’s public numbers being approximately correct is the minimum expectation, not a credit. But a claim that falls apart under the source data is worth nothing to anyone arguing against these systems in a council chamber, and it hands the other side an easy rebuttal.

Volunteers had already mapped 93 percent of it

There is a third count, and it was not produced by the company or extracted from its servers.

Documented on the map right now are 115,062 active Flock-brand ALPR cameras, each one recorded by a volunteer who physically saw it and logged it. Set against the 123,638 cameras Flock’s own records list as in service, residents with phones have independently accounted for roughly 93 percent of the company’s deployed camera fleet, with no access to any vendor system.

Across all ALPR brands the documented total is 142,848 active cameras, with a further 10,179 recorded and later observed as removed, for 153,027 ever documented. None of the researcher’s data has been merged into that map, and none of it will be. The value of a street-level record is that its provenance is known and independent, and that independence is exactly what allowed it to serve as a check on the vendor’s numbers here.

The network is not cameras

The genuine finding in the dataset is composition. Counting only devices in service, the network includes:

• 118,908 Falcon plate reader cameras, plus 2,511 Flex and 2,219 Highway variants
• 34,031 Wing devices
• 21,817 Picard processing units
• 15,807 Raven acoustic detection units
• 13,231 Condor devices
• 725 TalkDown units, which allow an operator to speak through the device
• 84 drone and drone infrastructure devices, including docking stations and radar

Nearly 16,000 in-service acoustic sensors is not a detail. Audio detection raises a distinct set of questions from plate reading, and those questions are frequently never asked locally, because a council that argued about cameras often never learned that microphones were part of the same purchase. The same applies to devices that can broadcast a voice onto a street, and to drone infrastructure sitting inside what is commonly described to the public as a camera contract.

83,427 devices that have not been built yet

The planning records deserve more attention than they have received. Across all device types, 83,427 rows are marked as in planning, against 214,073 in service. That is a pipeline equivalent to roughly 39 percent of the currently deployed network.

Planned is not the same as approved, funded, or installed, and some of those records will never become hardware. But the figure describes intended growth rather than growth already achieved, and it is the number most relevant to anyone deciding whether to act now or later.

The takedown demand is about trademark, not accuracy

In September 2026, Michael received a complaint from Doppel, a firm describing itself as an AI-native social engineering defense platform, saying it was acting on Flock’s behalf. The demand was that he stop using the FLOCK SAFETY mark and take the site down, on the theory that it causes customer confusion.

The objection is to a brand name, not to the correctness of the coordinates. Nothing in the complaint, as reported, asserts that the locations are wrong. The Intercept visited six randomly chosen Arizona locations from the dataset and found a Flock camera at each. Michael has declined to remove the map and has added a disclaimer stating the site is not affiliated with or endorsed by Flock.

That choice of instrument is telling. A company confident the data was inaccurate would say so. Objecting to the label on the page instead is a decision about which argument to have.

Why this matters locally

National totals settle nothing. Nobody votes on 335,701 rows. People vote on the agreement in front of their own city council, and the practical use of this dataset is the questions it makes specific.

Ask the agency how many devices it operates and how many are cameras. Ask which models. Ask whether any of them detect audio, and whether any can transmit sound. Ask how many additional devices are planned under the current agreement. Ask when it comes up for renewal. Those are answerable questions, and an agency that cannot answer them has answered something anyway.

Look up what is documented near you on the map, read how the systems work on the learn page, and find the steps and the language for asking on the take action page.

Note on this article

An earlier version of this piece, briefly published on September 27, 2026, carried the headline figure of roughly 170,000 cameras drawn from secondary reporting, and framed it as contradicting Flock’s public count. Checking the researcher’s published dataset directly showed that figure includes planned and decommissioned devices, and that the in-service camera count is 123,638. The article was corrected within the hour. The error was in relying on coverage rather than the source file, and the correction is noted here rather than made silently.

Sources

• The Intercept: Flock Wants the Most Detailed Map of Its Surveillance Cameras Taken Offline
• Tom’s Hardware: Flock seeks to have security researchers’ map of Flock cameras taken down
• Flock Surveillance Map, by Joshua Michael. Device counts in this article were computed from the dataset published on that site.