What problem does it solve?
Some fires start where a 911 call would be a long time coming. John Truett, fire management officer at the Arizona Department of Forestry and Fire Management, told the Associated Press that earlier detection means "we can launch aircraft and personnel to it and keep those fires as small as we can".
Brent Pascua, a battalion chief at the California Department of Forestry and Fire Protection (Cal Fire), said in the same report that for the fires where no 911 call would come for a long time it is extremely helpful to have the AI always monitoring the camera, and that in many cases his service started a response before 911 was even called. Neal Driscoll, founder of ALERTCalifornia, said the AI on the cameras is "actually beating 911 calls".
Utilities have a direct interest. Xcel Energy says AI cameras help protect communities, natural resources and power infrastructure, and that they build on measures it already runs, such as power line inspections, vegetation management and aerial surveillance. Arizona Public Service says its AI cameras have detected smoke starts within minutes of initially burning. A person cannot watch every ridge continuously, so software scans the images from each camera and a person decides what is real.
How does it work?
- Install cameras on existing structures. High definition cameras go on cell towers and other tall existing structures in high risk areas. Xcel Energy says each station sweeps 360 degrees every minute.
- Scan for smoke. A computer vision model checks every sweep for the shape and movement of smoke and must tell it from look alikes. APS says its system studies the way smoke looks to avoid mistaking other images like fog or dust for wildfire, and in AP's account of an Arizona detection the analysts checked that the smoke was not a cloud or dust.
- Verify with a person. An analyst at the vendor's intelligence center reviews each possible detection before an alert goes out. Xcel Energy describes this step, AP's account of an APS camera detection says human analysts verified it was not a cloud or dust, and the founder of ALERTCalifornia says human intervention keeps false positives low and trains the AI.
- Locate and alert. Two or more cameras triangulate the smoke, and the system sends the location and live video to the utility's fire team, the local dispatch center and fire agencies.
- Respond and learn. Agencies launch ground or air response, the utility decides whether to change operations near its lines, and the outcome is fed back to improve the model.
- Audience
- Employee facing
- Autonomy
- Copilot
- Adoption
- Early adopters
- Channels
- Internal tools, API and system to system
What is it worth?
Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.
No public deployment has disclosed a measurable outcome yet.
Value drivers: Risk and loss reduction, Speed and cycle time.
Indicative value
A utility that deploys 40 wildfire detection cameras
USD 1.5 million to USD 3 million
Annual cost of the detection service per year
How this is calculated
Formula: cameras * pricePerCamera. The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| Cameras in service cameras, cameras | 30 | 60 | Editorial assumption. Arizona Public Service had nearly 40 active cameras in May 2026 and planned 71 by the end of that summer, according to the Associated Press. |
| Annual service price per camera pricePerCamera, USD per camera per year | 50,000 | 50,000 | The Associated Press reports that Pano AI charges around USD 50,000 a year per camera, including fire risk analysis and a 24 hour intelligence center. Other vendors may price differently. Source |
What it leaves out: This is the cost side only. The value of earlier detection is avoided fire damage, liability, outage time and harm to people, and no public source in this library gives a per fire figure for a utility. Compare the cost with your own estimate of avoided losses, and add installation, power and connectivity costs.
Who already uses it?
3 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.
Arizona Public Service (APS)
United States · Energy and utilities · 2025
Arizona Public Service operates Pano AI cameras that watch 24 hours a day for early signs of wildfire in high risk areas and alert APS fire mitigation staff and fire dispatch centers when they detect smoke and heat traces, with dispatch centers given live feeds.
APS reported in 2025 that a camera on a communication tower in the Prescott National Forest detected smoke from the Brady Fire when it was less than one acre, and that the fire did not affect its equipment or service.
In May 2026 an Associated Press report said APS had nearly 40 active cameras and planned 71 by the end of summer, and quoted an APS meteorologist saying the cameras notify the utility about 45 minutes faster on average than the first 911 call.
No outcome disclosed.
Portland General Electric (PGE)
United States · Energy and utilities · 2025
Portland General Electric began installing AI wildfire detection cameras in 2021 and by July 2025 ran almost 40 of them around Portland and surrounding areas. When the cameras detect smoke or fire activity, agency users get a notification with the location and a link to the live feed. More than 215 users from 50 agencies use the network, and PGE opened the feeds to the public in July 2025. PGE's wildfire planning and analytics manager wrote in 2023 that in summer 2022 the cameras detected a fire south of Hagg Lake 140 minutes before traditional 911 and satellite detection, so Oregon Department of Forestry crews could drive straight to it.
No outcome disclosed.
Xcel Energy
United States · Energy and utilities · 2024
Xcel Energy installs Pano AI camera stations on towers and other tall structures near its power lines. Each station sweeps 360 degrees every minute, AI looks for smoke, Pano analysts verify a possible fire, and the system triangulates the location and notifies local fire agencies and dispatch centers.
In July 2024 Xcel and CORE Electric Cooperative described the Bear Creek Fire in Colorado: the Pano system and the US Forest Service detected the smoke from a lightning strike at the same time, a helicopter crew launched when the fire was 0.25 acres, and the fire was contained at three acres. At that date Xcel had 28 cameras working in Colorado.
It has since announced programmes for more than 50 camera stations in the Texas Panhandle (December 2024) and 38 in Minnesota (October 2025). In May 2026 the Associated Press wrote that Xcel Energy in Colorado "has installed 126" cameras and aims to have cameras in seven of the eight states it serves by year's end.
No outcome disclosed.
How do you implement it?
A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.
Data you need
- A map of high risk areas, line routes and existing towers or structures
- Terrain and viewshed analysis to place cameras with overlapping coverage
- Contacts and alert procedures for every fire agency and dispatch center in the territory
- A record of past ignitions and 911 call times to compare detection time
Systems to integrate
- Camera vendor platform and its intelligence center
- Fire dispatch and emergency management systems
- Utility operations center and outage management system
- Weather stations and fire risk models
Complexity: Medium
Cameras and models are supplied as a managed service. The utility needs sites with power and connectivity, usually on towers it does not own, permits, agreements with fire dispatch centers on how alerts are received, and a decision on who acts on an alert. Coverage needs overlapping cameras: Xcel Energy notes that accuracy improves as the stations' reach overlaps.
- 1
Choose sites by risk and coverage
Place cameras where ignitions near lines are most likely and where two cameras can see the same ground, so smoke can be triangulated. Xcel Energy's Bear Creek detection used two stations, one installed for CORE Electric and one for Xcel Energy.
- 2
Agree the alert path with agencies
Decide who receives a verified alert, by what channel and how quickly, and give dispatch centers and fire agencies access to live feeds. APS gave dispatch centers access to live camera feeds and fire detection information.
- 3
Run a season before counting results
Log every alert with its time, verification result and outcome, and compare the time to the first 911 call or other detection. Count false alarms and missed fires found by other means.
- 4
Tune verification and thresholds
Review false alarm causes with the vendor, adjust where cameras point and the alert rules, and record the changes.
- 5
Connect to operations
Decide how a confirmed fire near a line affects grid operations, and keep any shutoff decision with a person in the control room.
- 6
Share and expand
Open feeds to more agencies once alerts are reliable, and add cameras to gaps found in coverage reviews. PGE shares its network with more than 215 users from 50 agencies.
Guardrails
- A person verifies each detection before an alert goes to dispatch, at least until false alarm rates are known
- The camera system never operates the grid; any decision to shut off power stays with the control room
- Cameras are aimed and masked to avoid recording private property and people beyond what fire detection needs
- Alert, verification and response times are logged for every event
KPIs to instrument
- Minutes from first smoke to verified alert, and to first dispatch
- Minutes ahead of the first 911 call or other detection, where there is one
- Verified detections per season and false alarms per verified detection
- Fire size at first response
- Share of high risk territory covered by two or more cameras
Human in the loop
Vendor analysts verify possible detections, dispatchers decide how to respond, and fire commanders direct suppression. Utility fire mitigation staff decide what the alert means for operations near lines and review each season's alerts, false alarms and response times with agencies.
Common failure modes
- Alert fatigue
- Too many false alarms make dispatchers ignore alerts. Keep a human verification step and track false alarms per detection.
- Coverage gaps
- Terrain hides smoke from a single camera, and a fire starts in the gap. Plan overlapping coverage and review blind spots every season.
- Detection without response
- An early alert nobody acts on changes nothing. Agree the alert path and response times with agencies before the cameras go live.
- Unverified time savings
- A claimed lead time over 911 is easy to overstate, because it depends on which fires are counted. Record one comparison method and publish the sample size.
What are the risks and rules?
EU AI Act
Depends on design
Annex III point 2 makes AI systems high risk when they are intended to be used as safety components in the management and operation of the supply of electricity.
Recital 55 describes safety components as "systems used to directly protect the physical integrity of critical infrastructure or the health and safety of persons and property but which are not necessary in order for the system to function", and gives "systems for monitoring water pressure or fire alarm controlling systems" as examples.
Monitoring and alarm systems that do not switch anything can therefore fall inside the term, and having a person decide after an alert does not by itself remove a system from Annex III. The tier turns on whether the utility uses the cameras as a safety component in operating electricity supply, for example to protect lines, people or property, or only as a situational awareness tool for fire agencies.
A provider that concludes the system is not high risk must rely on the Article 6(3) derogation, which requires the system not to pose a significant risk of harm to health, safety or fundamental rights, and must document that assessment before the system is put into service. A design in which a detection triggers protective shutoff or reclosing settings automatically is the clearest case for high risk treatment. Cameras that record people or private property also bring GDPR duties.
Guidance
- Annex III, high risk AI systems referred to in Article 6(2) (European Union, Europe). Point 2 lists safety components in the management and operation of critical infrastructure, including electricity supply.
- Recital 55, safety components of critical infrastructure (European Union, Europe). Describes safety components as systems that directly protect the physical integrity of critical infrastructure or the health and safety of persons and property, with monitoring and fire alarm systems as examples. Whether the cameras are used that way decides the tier here.
- Article 6, classification rules for high risk AI systems (European Union, Europe). Paragraph 3 lets a provider treat an Annex III system as not high risk only after documenting its assessment before the system is put into service.
- AI Risk Management Framework (National Institute of Standards and Technology, North America). Voluntary US framework for mapping, measuring and managing AI risk, used by utilities that run the cameras in the United States.
Controls to put in place
- Written alert procedure signed with each fire agency, with named contacts
- Logs of every alert, verification decision and response time
- Seasonal review of false alarms, missed fires and coverage gaps
- Privacy review of camera placement and recording
Frequently asked questions
- How much earlier do AI cameras detect wildfires?
Two utilities report figures. Arizona Public Service says its cameras notify it about 45 minutes faster on average than the first 911 call, as quoted by the Associated Press in May 2026. A Portland General Electric manager wrote in 2023 that in summer 2022 the cameras detected one fire south of Hagg Lake 140 minutes before traditional 911 and satellite detection. The first is an average with no stated sample, and the second is a single fire, so neither predicts your own lead time.
- Does a person check every detection?
The utilities on this page describe a human step. Xcel Energy says analysts at Pano verify a potential fire before the system triangulates its location and notifies fire agencies, and AP's account of an APS camera detection says human analysts verified it was not a cloud or dust before alerting the state forest service and the utility. Keep the step until your own false alarm rate is known.
- How much do AI wildfire cameras cost?
The Associated Press reports that Pano AI charges around USD 50,000 a year per camera, including fire risk analysis and a 24 hour intelligence center. Cost is described as one of the biggest obstacles to adoption, and the report notes that false alarms are costly in time and attention.
How to cite this page
Blits.ai AI Use Case Library, "AI wildfire smoke detection cameras for utilities", last verified 13 October 2026, https://www.blits.ai/ai-use-cases/wildfire-smoke-detection-cameras. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 13 October 2026: First published