Big Story: How Two Cities Are Deploying AI Cameras Responsibly
Key Takeaways
Cleveland and Dallas have equipped city vehicles with cameras paired with AI vision-language technology to detect property conditions such as abandoned cars, high grass, graffiti, and illegal dumping.
Vision-language models can answer a wide range of questions about a single image without requiring a separate system for each use case.
Cleveland's survey had covered roughly 158,000 parcels by May, which under the prior process, would have required 40 officers working for 6 months.
Dallas identified nearly 29,000 potential property maintenance concerns across 49 issue types and roughly 12,900 right-of-way encroachment issues in its initial rollout.
Both cities built privacy protections into the programs, limiting imaging to conditions visible from public right-of-way and requiring human review before any enforcement action..
Cities have long struggled with similar problems. A porch collapses, tires pile up on a vacant lot, grass grows untended, and the condition often persists for weeks before anyone in city government learns of it, let alone responds. Cleveland and Dallas are now testing a technology that changes how quickly that discovery happens. Both cities have mounted cameras on vehicles driving on city streets and paired the footage with AI vision-language software, giving code enforcement teams a continuous visual record of property conditions across entire jurisdictions.
Cleveland began its partnership with City Detect in August 2025. Cleveland has roughly 18,000 vacant lots citywide, a volume that no reasonable increase in inspector headcount could keep pace with under the manual system. Dallas faced a similar constraint working across nearly 400 square miles. The city mounted its cameras on sanitation brush trucks, which covered wide sections of the city on regular routes. Brita Andercheck, Dallas's chief data officer, has said her team initially raised a concern that if cameras identify far more violations than inspectors can process, the result is a longer backlog.
A compliance letter alone results in roughly 80% of property owners correcting the issue voluntarily. If most owners fix a problem once notified, the limiting resource in code enforcement is how quickly the city learns a condition exists and how fast it can notify the owner.
The underlying technology differs from prior generations of municipal computer vision. Earlier systems answered one question each. Pavement-scoring tools rated road surface condition, and license-plate readers matched plates against a database. Each new use case required a new system. Vision-language models instead describe and reason over an image using an open vocabulary, so a city can pose new questions to the imagery it has collected without retraining a dedicated model. The same drive down a street can support code enforcement, 311 service requests, public works assessments, and storm damage review.
In practice, neither city has removed humans from enforcement. Cleveland officials review flagged images before anyone drives out to confirm a violation, and the city issues no citation based on a camera image alone. Dallas validates every AI-generated detection against staff judgment before any enforcement action proceeds, according to Code Compliance Director Chris Christian.
Privacy safeguards were built into both programs from the start. Dallas and Cleveland limit imaging to what is visible from the public right-of-way, covering parcels, pavement, and structures. Cleveland's approach follows Ohio open records and code enforcement laws, which restrict photography to the right-of-way. Dallas goes further on data handling, requiring its vendor to delete any image that shows no violation.
Cleveland mapped its grass-cutting process from complaint to resolution before looking for technology to address it. That problem, first sequencing, paired with clear rules on what imagery may be collected and how long it is kept, offers a template other jurisdictions weighing AI vision tools for property inspection, public works, or code compliance may find useful as they build their own governance frameworks around similar systems.
Build the right capabilities for what’s coming. Book a quick call with the Fractional Source team.
Quick Hit News:
A Utah State Board of Education investigation tested 100 school applications and found that, among 85 apps with standard data privacy agreements, 52% collected at least one type of data their agreements did not permit. 13% shared unpermitted data with third parties, while 36% shared data with advertising-related entities. The findings have since reached state lawmakers.
Los Angeles Unified School District has blocked student access to generative AI on district-issued devices while it develops a formal policy, a shift from its prior rule that allowed students 13 and older to use approved AI tools. The restriction took effect this school year but went largely unpublicized until this week, and board member Nick Melvoin raised concerns that the change was not explicitly approved by the board. The block relies on existing web filtering software, which can miss embedded AI features inside other approved platforms.
Meta has opened its $1.2 billion, 750,000-square-foot data center in Kuna, Idaho, its first in the state and 28th in the U.S. The facility is only the second facility to use a closed-loop cooling system, which recirculates roughly 200,000 gallons of water and can operate for 7 to 10 years before needing to be refilled, reducing the need for continuous cooling water. Meta is also backing three solar projects expected to add 645 megawatts of capacity to Idaho Power’s grid.
The Decatur City Council has approved new rules requiring any future data center in the city to obtain a special exemption permit through a multi-step review process that includes studies and a public comment period. No developers are currently pursuing a project there, but leaders wanted regulations in place before one arrives. Statewide, Alabama's Commerce Department reports $2.4 billion in confirmed data center investment, though independent research estimates the actual figure, including proposed and under-construction projects, at more than $63 billion.
For the Commute:
Infrastructure, Emergency Management, and ELGL Conference Preview with Chase Bruton (GovLove Podcast)
Chase Bruton, Town Manager for Yorktown, Indiana, discusses how the town handled recent flooding and what that response revealed about local emergency management. He also walks through upcoming infrastructure projects and the planning challenges facing a growing community. The conversation closes with a preview of the ELGL Gov Grand Prix Conference and what attendees can expect in Indianapolis.
Resources & Events:
📅 Governing AI Together (Virtual - September 17, 2026)
Government Technology and the Center for Digital Government will bring together public-sector finance, IT, security, and operations leaders for a webinar on building shared governance structures for AI adoption. The session will cover decision rights, sensitive-data protection, compliance, risk management, departmental alignment, and practical guardrails for scaling AI responsibly across government organizations. Details →
📅 Georgia Emerging Technology Summit (Atlanta, GA - November 12, 2026)
This summit brings state, regional, and local public sector leaders together to explore AI, automation, edge computing, and data governance through demos, policy discussion, and case studies. Georgia Chief Digital and AI Officer Nikhil Deshpande and Georgia Technology Authority Executive Director Shawnzia Thomas will speak, alongside concurrent workshops on AI governance, agent management, and workforce training. Details →
📊 Report Spotlight: 2026 State CIO Top 10 Priorities (NASCIO)
NASCIO's 20th annual survey of state and territory chief information officers shows artificial intelligence rising to the top priority for 2026, ending cybersecurity's 12-year run at the number 1 spot. Cybersecurity holds steady at 2nd place, while budget and fiscal management climbed to 3rd from 6th the year before. Accessibility jumped 4 spots after making its first appearance on the list last year, a shift the report ties to new federal web accessibility rules taking effect for larger jurisdictions. Read →
Insight of the Week:
The University of Chicago’s social sciences division has issued guidance to make its core courses largely analog, asking students to leave digital devices out of class and prohibiting the use of AI for coursework. The policy is designed around small, discussion-heavy classes of no more than 20 students, where faculty want students to focus on reading, writing, and face-to-face argument. The guidance also discourages instructors from outsourcing grading to teaching assistants or large language models, although limited exceptions remain for accessibility, database work, and carefully validated AI-assisted grading.
Was this email forwarded to you? Subscribe here to get Fractional Leader delivered to your inbox every Tuesday & Thursday.
