California police surveillance technology spending
Explore spending by department across California. Each bubble is one department. Bubble size reflects the metric you choose.
Developed by the Alliance
Share data with usDiscussion
In this discussion, we address three questions: how much are these departments spending on surveillance technology, what are they buying, and what best explains who spends the most?
A note on scope: the figures here come from the contracts that departments returned in response to the records requests we sent to the 100 most populous California cities, which asked for surveillance-technology agreements dated from the start of 2024 onward. 47 departments returned relevant contracts. Because many are multi-year deals, we spread each contract’s cost month-by-month across the years it is active. In this discussion, we focus on 2024 and 2025.
On the map, total cost is spending in each calendar year; annualized cost sums the average yearly cost of contracts active in the selected time period. In this discussion, dollar totals are total cost unless noted otherwise.
1. How much are police spending?
Total spending rose from about $5.5 million in 2024 to $8.1 million in 2025, an increase of roughly 50 percent.
That increase is likely overstated. Because the records request only covered 2024 onward, departments may have excluded contracts that were signed before 2024 but remained active that year, which would exaggerate the apparent increase. At the most conservative extreme, total spending was nearly flat across the few departments whose files include pre-2024 contracts.
A typical department’s annualized cost was about $803 per 1,000 residents in 2024 and $1,147 in 2025, but the spread is wide: most departments spend modestly while a handful spend many times more. Each dot is one department.
2. What are they buying?
We divided total spending into eight categories of surveillance technology. Of these categories, one accounts for most of the spending.
License plate readers are pole- and car-mounted cameras (e.g. Flock) that photograph every passing plate and build a searchable history of where cars have been. They are the largest area of total spending, about $3.6 million in 2024 and $4.4 million in 2025.
Gunshot detection uses acoustic sensors that try to pinpoint gunfire (e.g. ShotSpotter, now SoundThinking). It is the second-largest category by total spending, rising from about $819 thousand in 2024 to $1.6 million in 2025.
Real-time crime centers are software platforms that pull many camera and sensor feeds into one dashboard for officers (e.g. Fusus, Flock ARTIC). Total spending rose from about $311 thousand to $552 thousand.
Data analysis platforms aggregate and analyze records from multiple sources for investigations (e.g. Peregrine, CrimeTracer, ForceMetrics). Total spending rose from about $267 thousand to $398 thousand.
Other surveillance cameras are fixed and mobile video cameras not tied to plate reading, such as pan-tilt-zoom units and trailer-mounted camera towers (e.g. Flock Condor). Total spending on them rose from about $306 thousand to $547 thousand.
Drones are unmanned aircraft used for overhead surveillance, increasingly dispatched as the first responder to a call (e.g. Flock Aerodome, Axon Air). Total spending grew from roughly $50 thousand to $286 thousand.
Generative artificial intelligence refers mostly to tools that draft police reports automatically from body-camera audio (e.g. Axon’s Draft One and Veritone). Total spending grew from roughly $56 thousand to $267 thousand.
Facial recognition and biometrics are tools that try to identify people from their faces or other physical features (e.g. Clearview AI, AFR Engine). They are the smallest category by total spending, under $100,000 across both years.
3. What explains who spends the most?
We tested whether total spending tracks a city’s crime, its wealth, its demographic makeup, or its size. We could not find a reliable link to any of them.1
We did find that total spending lines up with grant money. In late 2023 California announced its largest-ever investment to fight retail theft — about $270 million to local police through the Organized Retail Theft Prevention grants. Departments that won a grant had roughly three times the median annualized cost per 1,000 residents in 2024 as those that didn’t (about $1,156 versus $418).23
This spending pattern suggests that grant money to prevent retail theft is often spent on general-purpose surveillance infrastructure such as automated license plate readers. A review of all 38 funded applications corroborates this: almost every one (37 of 38) proposed automated license plate readers, about a third also proposed real-time crime centers or gunshot detection, a smaller share proposed data analysis platforms, and a smaller share still proposed facial recognition or biometrics. The proposed plate-reader counts were often large: Fresno budgeted for 50, National City for 94, the Ventura County Sheriff for 100, and Modesto for 200.
Notes
- A city’s wealth, racial makeup, population-adjusted size, and its neighbors’ total spending all failed to line up with total spending in either year. A combined model that includes funding, crime, demographics, and department size together explains only about a tenth of the variation and produces no statistically reliable predictor.
- This is a simple comparison of grant recipients with everyone else. The quantitative relationship between a department’s exact grant dollars and its total surveillance spending is not statistically reliable in this sample.
- These contracts are almost certainly an incomplete record of what departments bought. The clearest sign is that the dollars do not reconcile: the departments that won retail-theft grants received about $61.8 million between them, yet the surveillance contracts they disclosed account for only a small fraction of that money. Some of that gap is expected, since the grants also fund things outside these surveillance categories and are spent over several years, but a shortfall this large suggests that some contracts were not produced in response to the records requests.
A note on these figures. Full collection and extraction details are in the Methodology section below. For the comparisons in this discussion, population is from the United States Census Bureau, crime from the Federal Bureau of Investigation, and grant awards from California’s Board of State and Community Corrections. Five places (Compton, Lake Forest, San Marcos, Thousand Oaks, and Victorville) are policed by county sheriffs rather than their own departments; they are counted as buyers but left out of comparisons that depend on a department’s own size.
Methodology
Data
Data was compiled from CPRA requests to California police and sheriff departments. These requests were issued by members of the Alliance to the 100 most populous cities in California.
Members used the following template.
Subject: Public Records Act Request: AI and Automated Decision Systems Procurement
Pursuant to the California Public Records Act (Gov. Code § 7920.000 et seq.), I am requesting electronic copies of any finalized Contracts, Data Sharing Agreements (DSAs), Requests for Proposals (RFPs), and Acceptable Use Policies dated from January 1, 2024, to the present, regarding the procurement, testing, or use of the following technologies by the municipality or its police department:
- Biometric identification and facial recognition systems (e.g., Clearview AI).
- Predictive policing software, acoustic detection, or Real-Time Crime Center integrations (e.g., SoundThinking/ShotSpotter, Fusus).
- Automated license plate readers (e.g., Flock Safety).
- Generative AI models or Automated Decision Systems (ADS) used to process citizen data, automate investigations, or draft official police narratives (e.g., Axon Draft One, OpenAI).
If you believe this request is too broad, I invoke my rights under Gov. Code § 7922.600 and request that you assist me in identifying the specific records and information technology systems that hold these contracts.
Please provide these records in electronic format. Because I am requesting electronic copies of existing documents, there should be no cost of duplication. If you anticipate any fees for fulfilling this request, please contact me for authorization before proceeding.
Despite the fact that we only requested documents from 2024 and later, we received documents from earlier years from some departments. Therefore, if you are planning to use cross-department data for professional purposes, we recommend only using data from 2024 and later.
Reporting also suggests our data is incomplete; for example, the San Francisco Standard reported in June 2025 that San Francisco Police Department (SFPD) was piloting Axon Draft One, but SFPD indicated they had no documents to release in response to our request.
Request the raw data (documents provided by the police and sheriff departments) by emailing contact@worldalliance.org.
Analysis
We used AI (Claude Opus 4.8) to process the documents and extract line items from them for each department. We developed a steering document that we used to guide the AI's extraction. Before we ran the automated extraction, we manually reviewed a sample of the extracted line items to ensure the AI was extracting the correct information. We also manually checked a large number of line items after extraction to verify accuracy.
We do not include repair costs. Many contracts lump repairs in with hardware or service fees which made it difficult or impossible to determine what was being repaired, so we excluded those line items from the dataset.
Annualized cost is calculated as total contract cost divided by contract length in months (times twelve). Some contracts listed payment schedules, but we used a uniform approach because most contracts did not. In most cases, the difference is less than 5%.
Credit
Contributors: Nihar Doshi, Sidney Hough, and Mark Xu.