13,000 cameras and 615,000 searches: What exactly has Texas built?

Start with the numbers

Texas does not have 615,000 Flock cameras. What it has may still surprise you.

A Texas Tribune review found roughly 615,000 searches of the Texas Department of Public Safety Flock network during June 2026—about one search every four seconds. Those searches came from 3,075 law-enforcement agencies. Forty-eight percent came from outside Texas. Houston Police accounted for 13 percent of the searches, while DPS itself accounted for just under nine percent.

The exact statewide camera count is unknown because neither DPS nor Flock publishes a complete inventory. The Tribune reported that DeFlock, a crowdsourced anti-surveillance mapping project, had identified about 13,000 cameras in Texas. State funding tied to a $1 increase in Texans’ auto-insurance fees financed at least 3,200 Flock cameras. As of early September, public information indicated at least 940 DPS cameras had been installed.

That distinction matters. A camera count tells us the physical footprint. Search volume tells us how heavily the accumulated data is being queried. Both deserve scrutiny, but confusing them exaggerates the hardware footprint by almost fifty times.

What the cameras collect today

This is already more capable than a simple license-plate reader.

Flock's system uses machine learning to identify plates and vehicle characteristics. Reporting based on DPS records describes searches by make, model, color, bumper stickers, bike racks and other characteristics. The system can also identify vehicles that frequently travel together. Flock describes its ALPR product as vehicle-focused and says it does not use facial recognition, identify drivers or passengers, collect demographic information, or perform predictive policing.

Those statements should not be casually converted into “Flock secretly recognizes faces.” We found no evidence that the current ALPR product does that.

But absence of facial recognition does not mean absence of tracking capability. A sufficiently dense sequence of timestamped vehicle observations can reveal where a particular vehicle appeared and how it moved through the camera network. A license plate also frequently provides an indirect path to a registered owner through authorized databases. That makes vehicle-location history sensitive even without biometric identification.

Could the capability change?

The network is software-defined—but that does not make every future feature possible on today's hardware.

Flock's own FAQ says its subscription includes software updates and describes the company as “just as much a software company as we are a hardware company.” It says camera software can be updated in real time when new features are released, and its current product terms provide customers access to generally available enhancements and new features within their product configuration.

That establishes an important governance fact: the capabilities of deployed devices are not necessarily frozen on installation day.

It does not establish that facial recognition could simply be switched on tomorrow. Reliable facial recognition could require different optics, image composition, resolution, processing, data handling, product licensing and legal authorization. We found no technical evidence showing that existing Flock ALPR hardware can be transformed into a reliable facial-recognition network through software alone.

The better public-policy question is therefore broader: what process governs a material expansion in capability after thousands of devices have already been deployed? Who approves it? Must the public be notified? Does a new capability require a new contract, ordinance or warrant standard? Those controls matter precisely because software-defined infrastructure can evolve.

The access-control problem

The most revealing numbers may be about permissions, not cameras.

DPS told the Texas Tribune in July that agencies needed signed agreements to access its Flock data and said 206 Texas departments had such agreements. Yet the Tribune's June audit found 3,075 agencies had searched the network, including 2,679 outside Texas. After the newspaper questioned that discrepancy, DPS announced that only agencies with signed agreements would be allowed to conduct searches going forward.

More than 58,000 June searches were categorized as traffic infractions. Another 111 searches listed immigration-related details, including Homeland Security Investigations, Customs and Border Protection, and an “ICE deportation warrant.”

This is where an IAM/PAM perspective becomes useful. A surveillance system is not defined only by what its sensors can see. Its effective power is also determined by who can authenticate, what they are authorized to query, what other datasets can be correlated, whether access is federated across organizations, how activity is audited, and what happens when privileges are abused.

The expansion of access beyond what DPS had publicly described is not hypothetical. It is a documented example of permission scope becoming larger than the public understanding of the system.

Misuse is not hypothetical either

Audit logs matter because trusted users sometimes violate trust.

Houston-area agencies have investigated multiple alleged or confirmed cases of improper Flock use. Harris County suspended three deputies' access during an investigation. Deer Park disciplined an officer after an audit. Reporting has also described a Fort Bend County lieutenant admitting to nearly 190 improper searches and a Pasadena sergeant resigning after an investigation found he used the system to track a female officer.

Flock has responded with stronger controls. The company announced a seven-day default retention recommendation, required case codes, additional auditing, proactive lockouts and mandatory multi-factor authentication. Flock says every search is logged and tied to an individual user.

Those safeguards are meaningful. Their existence also illustrates the core point: protecting a powerful database depends on governance and access control, not merely on the intentions behind installing the cameras.

Where does all this data go?

The cloud connection is documented.

Flock's August 2026 information-security addendum states that the company exclusively uses Amazon Web Services for data-center facilities. Its LPR policy says data is briefly stored on the camera, encrypted and transferred to AWS U.S. cloud infrastructure; CJIS information is stored in AWS GovCloud. Flock's current default retention is seven days, although different retention may apply under law or customer policy, and its Evidence Mode can preserve selected records for active investigations.

So the architecture really does look like this:

Roadside sensor → cellular network → cloud infrastructure → machine-learning processing → searchable database → authenticated users → cross-agency sharing.

That is a cloud-dependent surveillance architecture. Calling it that does not require speculation.

Now follow the data centers

Texas's compute boom is enormous—and unusually opaque.

Reuters reported that requests from data centers and other large users to connect to the Texas grid grew from about 48 gigawatts in 2023 to more than 474 gigawatts in 2026. Texas responded by auditing proposed facilities and demanding more information about ownership, funding, water, generation and project viability. Regulators are also examining “ghost demand”—speculative projects that reserve enormous amounts of prospective grid capacity without necessarily being built.

Texas has separately begun enforcing water-use reporting against data centers. ERCOT's current request for information covers projects of 25 megawatts or larger and specifically asks about ownership and controlling interests, tax benefits, water, grid dependence, on-site generation, cooling, noise, light and traffic. A public report is expected by December 10, 2026.

Amazon itself is planning Texas data-center infrastructure, including a campus outside Boling in Wharton County. That fact creates an obvious question because Flock uses AWS. But sharing a cloud provider is not evidence that a particular Texas AWS facility stores Flock data, and it is certainly not evidence that Texas data centers are being constructed for surveillance.

The correlation question

We looked for the bridge. We did not find it.

We searched current reporting, Flock's security and product documentation, Texas data-center disclosures, state regulatory material and reporting about the Texas Flock network for evidence that the current hyperscale data-center expansion is being driven by, dedicated to, or coordinated with Flock-style surveillance.

We found a technological bridge: Flock relies on AWS data centers, and surveillance at this scale depends on cloud storage, networking, databases and machine-learning compute.

We found no evidence of the stronger bridge: no public contract, project announcement, state directive or technical disclosure showing that the Texas AI data-center boom is a covert buildout for the Flock network or a unified mass-surveillance program.

That distinction is essential. Two rapidly expanding infrastructures can complement one another without being parts of the same secret plan.

What deserves watching next

The legitimate concern is capability convergence.

The most consequential future scenario does not require a conspiracy. Sensors are becoming cheaper and more numerous. Cloud compute is becoming vastly larger. AI can make enormous collections of observations searchable and correlatable. Agencies can share data across organizational boundaries. Software can add capabilities after physical infrastructure has been installed.

Put those trends together and societies can acquire surveillance capabilities incrementally—one camera contract, one cloud service, one sharing agreement and one software feature at a time—without ever holding a single public debate about the combined system.

That is where oversight should concentrate: purpose limitation, warrants and legal standards, retention, cross-jurisdiction sharing, biometric restrictions, API and third-party access, least privilege, immutable audit trails, independent audits, public disclosure of capability changes, and meaningful consequences for misuse.

Sources reviewed

Read the receipts.

Texas Tribune — 615,000 DPS-network searches and cross-agency access →

Texas Tribune — camera estimates and state funding →

Texas Tribune — capabilities, sharing and statewide expansion →

Flock Safety — stated ALPR limits and facial-recognition policy →

Flock Safety — software updates and product capabilities →

Flock Safety — AWS infrastructure and security controls →

Flock Safety — cloud storage, retention and LPR policy →

Reuters — Texas data-center interconnection boom and audit →

ERCOT — 2026 Data Center Impact RFI →

State of Texas — data-center water reporting and enforcement →

Source review completed September 21, 2026. Texas's data-center audit is ongoing, and ERCOT currently expects to publish its impact report by December 10, 2026. This investigation should be updated when that report and additional ownership disclosures become available.