The next municipal surveillance debate may not arrive in a police department budget. It may arrive in the sanitation budget.
Dallas has equipped 50 brush-and-bulky-waste trucks with 100 AI-powered cameras that photograph property conditions as the trucks travel their normal
routes. The system, supplied by City Detect, uses computer vision to identify possible problems such as illegal dumping, high grass, debris, graffiti and dangerous structures.
In roughly four months, the cameras flagged more than 20,000 potential code concerns. About 3,700 courtesy notices went out, and a September 8 city memo reported 57 citations at 34 properties, issued after city employees reviewed the detections and code officers inspected in person.
The city pays $852,000 a year under a three-year contract valued at an estimated $2.556 million. The City Council approved it unanimously in December 2025.
That makes this an interesting AI story. It makes an even better local-government story, because Dallas did not have to build a new surveillance network. It already owned one. It was called the garbage fleet.
The fleet was already going past every house
This is the clever part of the technology, and the part municipal officials should think hardest about.
Sanitation vehicles already travel neighborhood streets on predictable schedules. The city already pays the drivers, buys the fuel and maintains the trucks. By attaching cameras and software to vehicles performing an existing function, a government can collect detailed visual information across enormous portions of the community without creating a new inspection workforce.
Dallas described the system exactly that way when it funded the program. Its FY 2025-26 budget said cameras on 50 sanitation trucks would act as an early-detection system, scanning public rights-of-way during normal routes so Code Compliance could prioritize inspections and Sanitation could improve operations. Operationally, that makes sense. A code inspector cannot drive every street every month looking for weeds, damaged structures and illegal dumping. A garbage truck already does.
AI converts those routes into something different: a recurring municipal sensor network. And the concept will not end with garbage trucks. Snowplows can scan pavement, buses can monitor curb conditions, parking vehicles can identify damaged signs, street sweepers can map potholes, and utility trucks can inventory trees, poles and structures. The same technology that helps a city find a broken sidewalk can, depending on what is collected and retained, also create a remarkably detailed historical record of what was visible at a particular address on a particular day.
So the hard governance question is not whether this technology can make city employees more efficient. Of course it can. The question is what rules should exist before an ordinary city vehicle quietly becomes a data-collection platform.
Dallas asked some of the right questions
One reason the Dallas case is useful is that council members asked unusually concrete questions before the cameras were deployed.
A November 2025 city memorandum explains that City Detect, not Dallas, configures the AI system according to operational requirements set by the city. The cameras record 1920-by-1080 images using fixed lenses with no zoom, and facial and license-plate blurring is enabled. City Detect maintains the data for the duration of the contract, and Dallas may download and retain it during the agreement and for 90 days after it ends. The data are stored in U.S.-based cloud storage.
Council members also asked whether the information could be obtained through open-records requests or subpoenas, and the city said those requests would follow ordinary Dallas legal and public-information procedures. They asked whether police could access the system, and Dallas said the Police Department would have no direct access, because the contract is between Sanitation, Code Compliance and City Detect.
Those are exactly the questions local governments should ask. But the answers also show why this technology deserves more attention than a standard equipment purchase. The government is not simply buying cameras. It is creating a dataset.
Datasets have a way of becoming useful for things nobody planned
Municipal technology frequently begins with a narrow purpose. Then another department notices that the information might help solve its problem.
Privacy and technology discussions call this function creep. A traffic camera becomes useful to police. A license-plate reader built to find stolen cars becomes useful in other investigations. A building database created for permitting becomes useful for enforcement. A sanitation-camera database created to identify weeds and dumping could someday be useful for entirely different questions.
Dallas currently says police do not have direct access. City Detect says its system is not connected to federal or law-enforcement databases, that faces and plates are blurred by default, that humans review detections before action is taken, that an independent SOC 2 audit verifies its data handling every year, and that it does not earn revenue based on how many detections lead to formal action.
Those are meaningful safeguards. But safeguards are not permanent restrictions. Policies, contracts, administrations and technology capabilities all change, and data that exist can attract uses nobody contemplated when the council approved the original contract. That is why governments should decide before deployment which secondary uses are prohibited, rather than merely describing what today’s department intends to do.
Even the privacy filter has already produced a warning
Vendor safeguards should not be treated as infallible. City Detect says it automatically blurs faces and license plates. Yet on September 23, NBC 5 Investigates reported that an image it obtained through an open-records request contained a license plate the automated process had failed to blur. The city said it had to add a black box over the plate before releasing the record. Jeremy Reed, who oversees the program for Code Compliance, said the vendor described the blurring process as about 99% effective.
That does not mean the privacy system is defective overall. It means what local governments should assume about any automated safeguard: it will occasionally fail.
The Dallas images also reveal a problem that blurring does not solve. A photograph can hide someone’s face while still showing clothing, body characteristics, a distinctive vehicle, the time of capture and the address where the person or vehicle appeared. NBC’s review found that those contextual details remained visible in some records. That may be perfectly appropriate for code enforcement. But cities should understand that anonymized does not always mean unidentifiable.
Human review does not answer the equity question
Dallas and City Detect emphasize that the AI does not automatically issue citations. A human employee reviews each detection, and Dallas says an in-person inspection occurs before a citation. That matters.
But “human in the loop” should not end the analysis. The AI still determines which properties get attention first, and enforcement resources are limited. If software disproportionately sends inspectors toward one part of a city, human employees may unintentionally reproduce the distribution created upstream by the algorithm.
The early numbers raise that question. Southern Dallas accounted for roughly 13,800 of the more than 20,000 photographed concerns. On the enforcement end, the city manager’s September 8 memo said 39 of the 57 citations, or 68%, were issued in Council Districts 4, 7 and 8 in Southern Dallas, which NBC 5 described as predominantly Black and Hispanic. Council Member Chad West, who has warned the program could function as a tax on the poor, told FOX 4 that about 70% of citations came from predominantly African American neighborhoods.
Those numbers do not, by themselves, prove algorithmic discrimination. Older housing stock may genuinely have more observable maintenance problems, income differences may affect property conditions, truck routes may not generate identical observation opportunities, and neighborhoods may differ in lot design and visibility. Reed has acknowledged that the cameras will likely capture more violations in Southern Dallas, where he says more violations exist. All of that is exactly why the city should measure the system. The relevant question is not whether an algorithm is technically “unbiased.” It is whether the complete enforcement pipeline produces fair and defensible outcomes:
Where were properties photographed?
What percentage were flagged?
What kinds of violations were identified?
How many resulted in notices?
How many survived human review?
How many became citations?
How many were incorrect?
How long did residents have to cure the problem?
How do those patterns compare with code enforcement before the AI system existed?
Without that analysis, governments are left arguing about intentions. With it, they can evaluate outcomes. Dallas council members are due a briefing on the program in October, which is the natural moment to ask.
A surprisingly important performance question
The most seductive municipal technology pitch is often “we can find more problems.” But finding more problems is not the same as solving more problems.
If AI identifies 20,000 potential violations, a city has not automatically improved 20,000 properties. It may instead have created 20,000 additional items that employees must review, prioritize, inspect, document and potentially enforce. That can be useful. It can also simply move the bottleneck.
A city evaluating technology like this should avoid measuring success solely by detections or notices generated. Better measures include compliance rates, hazards corrected, illegal dumping removed, inspector hours saved, repeat violations reduced, resident response time and cost per resolved problem. Otherwise an AI system can look spectacular on a dashboard because it discovered thousands more things for government to do. That is not necessarily efficiency. Sometimes it is automated workload creation.
What cities should ask before turning the fleet into sensors
Before putting AI cameras on sanitation trucks, buses, snowplows or other municipal vehicles, a city should be able to answer a few basic questions:
What exactly is being collected, and what is explicitly prohibited from being collected or analyzed?
How long is raw imagery retained, who owns it, and what happens to it when the contract ends?
Which departments can access the information, and are secondary uses prohibited by ordinance or policy, or merely by current practice?
What happens when automatic privacy protections fail?
Can residents obtain the image or data that triggered enforcement and meaningfully challenge an erroneous detection?
Will the city publish geographic and demographic outcome data so elected officials can see whether enforcement shifts disproportionately?
What metric determines whether the program is working: detections, notices, compliance, staff savings or improved neighborhood conditions?
Who audits the system after purchase, rather than relying exclusively on the vendor that sold it?
These are not arguments against the technology. They are procurement questions, records questions, management questions and due-process questions. In other words, municipal-government questions.
The CityNewsWire takeaway
The most important thing about Dallas’ AI garbage trucks is not that a computer can recognize tall grass. That technology will become ordinary.
The important development is that routine municipal infrastructure can now double as continuous observation infrastructure. A city may buy the system because it wants cleaner alleys. A public-works director may see staff efficiency. A council may see a relatively modest technology contract. But the decision quietly creates something larger: a new stream of government data about the physical community.
That deserves the same governance attention cities are finally beginning to give police cameras, automated license-plate readers and facial recognition. The next surveillance system may not look like surveillance when the council approves it. It may look like a garbage truck.
CityNewsWire action
If your community is considering, or already using, AI cameras on municipal vehicles, request the vendor contract, retention schedule, data-sharing policy, user-access logs, AI configuration criteria, privacy-impact assessment, monthly detection data, notice and citation outcomes, false-positive records, geographic enforcement data and any policies governing secondary use of the imagery.
Do not just ask what the camera can see. Ask what government can do with what the camera remembers.
Sources
Dallas City Council resolution approving the City Detect contract, December 10, 2025
City of Dallas, Update on the Use of AI-Powered Camera Technology, Finance Committee presentation, December 9, 2025
Dallas Morning News, Dallas approved AI-powered cameras for sanitation trucks. What to know
Dallas Free Press, City of Dallas to put AI cameras on garbage trucks to identify code violations (FY 2025-26 budget)
Dallas Morning News, AI flagged 20,000 possible code violations, only 57 were cited
Dallas Observer, AI-powered trash truck cameras in Dallas capture 20,000 images, yield low citations
NBC 5 Investigates, Dallas AI code enforcement cameras lead to first homeowner citations (September 2026)
NBC 5 Investigates, Photo raises questions about trash truck AI cameras and privacy (September 23, 2026)
FOX 4 Dallas-Fort Worth, AI-powered cameras on Dallas trash trucks result in almost 4,000 notices, 57 citations
City Detect, Responsible AI Strategy and San Francisco Adds City Detect to Its Emerging Technologies Vendor Pool