Flox Intelligence Wildlife AI Takes On Animal Collisions at Airports, Railways, and Roads

Flox Intelligence is expanding its wildlife AI beyond fixed deterrents, with 150 edge devices already operating across three continents. The systems identify nearby animals and select species-specific sounds intended to move them away from runways, railways, roads, and other infrastructure.

The conflict is not simply animals versus machines. Flox is betting that adaptive, local intelligence can outperform fences, sirens, lights, lasers, patrols, and repeated sounds. Those established approaches remain useful, but each has limits when wildlife changes routes or becomes accustomed to a deterrent.

The company is now opening a $10 million Series A, according to Flox funding details. It wants to grow from roughly 150 deployed units to several thousand. That plan turns an unusual conservation technology into a test of whether edge AI can become dependable infrastructure.

Flox Intelligence Wildlife AI Moves From Experiments to Infrastructure

The important change is not a new animal-recognition demo. Flox is trying to turn detection, interpretation, and deterrence into one operating system.

Flox emerged from work conducted at the KTH Royal Institute of Technology in Stockholm. Founder and CEO Sara Nozkova had previously spent 13 years working in transportation and autonomous vehicles. The company spun out of KTH in 2020.

Nozkova’s premise was that conventional deterrents often treat wildlife as one predictable category. A siren can make noise, a light can flash, and a fence can redirect movement. None of those tools necessarily understands which animal has arrived or how that animal is behaving.

Flox Edge, the company’s current hardware product, uses a motion sensor to wake the system. A camera then captures the scene, while an on-device model attempts to identify the animal. The unit selects an audio response intended for that species and situation.

This is edge AI, meaning the device processes information locally instead of sending every observation to a distant cloud service. Local processing matters beside isolated tracks, remote roads, or large airfields where connectivity can be unreliable. It also reduces the delay between detection and response.

The company says its systems can vary their signals over time. That variation addresses habituation, the process through which animals stop reacting to a repeated stimulus. A fixed recording may become background noise after enough exposure.

Nozkova summarized her design philosophy plainly: “I was convinced the answer was not a louder deterrent.” Her alternative is a system that changes its response based on the detected species and observed behavior.

The hardware can be attached to poles, trees, and other existing structures. Customers also receive software for viewing deployments, events, and conditions across a site. The platform can produce reports for regulatory compliance and biodiversity monitoring, according to the company.

Flox says approximately 150 devices are active across North America, Europe, and Australia. Reported locations include airport perimeters, rail corridors, roads, farms, parks, mining sites, urban areas, and electrical substations.

That footprint remains small beside national transport networks. However, it is broad enough to expose the devices to different species, weather conditions, and operating environments. Those encounters generate the field data Flox considers central to its advantage.

The company sells its physical units and charges an annual software and AI subscription after the first year. It also licenses software through original equipment manufacturer partnerships. Alstom is testing that approach by bringing Flox technology onto railway equipment.

Flox expects about $1.5 million in revenue for 2026, Nozkova told TechCrunch. The company has raised approximately $5 million through its pre-seed and seed financings. Its proposed Series A would finance a much larger deployment and expand the technology beyond current wildlife applications.

The round has not been announced as closed. Flox therefore has an operational system and paying customers, but its next expansion remains a financing objective. That distinction matters when evaluating its timetable.

Wildlife Collisions Make This More Than a Conservation Project

Flox is selling safety and operational continuity, with animal protection serving the same outcome rather than a separate mission.

Wildlife enters infrastructure designed around controlled movement. Aircraft follow runways, trains follow rails, and cars follow roads. Animals do not recognize those operating boundaries, and their routes can shift with weather, migration, food, construction, and human activity.

The scale is clearest in aviation. The latest FAA strike data records about 344,000 reported wildlife strikes involving United States civil aircraft between 1990 and 2025. Around 24,500 strikes occurred at 870 American airports during 2025.

The FAA estimates that these incidents cost United States civil aviation an average of approximately $272 million annually during that period. That estimate covers reported direct and other monetary losses, not every ecological or operational consequence.

Birds represented 96 percent of reported American civil-aircraft strikes. Yet treating the problem as generic bird control can obscure major differences between species. The FAA says waterfowl account for only 4 percent of strikes but 23 percent of damaging strikes.

Most events also occur close to the ground. About 70 percent happen between ground level and 500 feet. That concentration places airport grounds and their immediate surroundings at the center of mitigation efforts.

The danger is not theoretical. A United Airlines Boeing 737 MAX 9 struck a coyote during takeoff from Chicago O’Hare in January 2025. The aircraft returned safely, but the O’Hare coyote strike forced an inspection and passenger disruption.

Railways face a related problem across much longer, less controlled corridors. Alstom says Sweden records approximately 5,000 animal collisions annually. These incidents can injure wildlife, damage trains, delay passengers, and expose drivers to traumatic encounters.

France offers another measure of the operational burden. SNCF says wandering animals affect more than 8,000 trains nationally each year. In Brittany, 120 animal collisions during 2023 disrupted 660 services and caused over 15,000 minutes of delays.

Fences remain important, especially near runways and high-risk rail sections. However, continuous fencing is expensive and can fragment habitats. Openings, crossings, damaged sections, and fence ends can also become predictable points of entry.

Human patrols bring judgment but cannot observe every remote corridor continuously. Trained dogs, falcons, pyrotechnics, vehicles, lights, and sirens can disperse animals. Their effectiveness depends on staffing, species, timing, terrain, and repeated follow-up.

That creates pressure on airport operators, rail companies, road agencies, utilities, and land managers. They need to reduce collision risk without depending entirely on labor-intensive patrols or lethal control. They also need evidence that any automated response remains effective after deployment.

Flox Intelligence wildlife AI targets that gap. It promises continuous observation, species classification, and a targeted response in one field unit. The company must now show that this integration improves outcomes beyond selected trials.

The value proposition also extends past individual collisions. Every recorded encounter can reveal where animals enter a site, when movement peaks, and which interventions change behavior. Infrastructure operators could use those patterns to improve fencing, schedules, inspections, and habitat management.

That data layer may ultimately prove as important as the sound itself. A deterrent reacts to one animal, while a mapped history can influence how an entire site operates. Flox is therefore competing for a place inside safety planning, not merely a place beside a runway.

Species-Specific Sounds Are the Core Mechanism

Flox’s central wager is that wildlife deterrence improves when a device identifies the animal before deciding how to respond.

The system begins with an event, not a continuous cloud video stream. Movement activates the camera and local model. The model classifies what it sees, after which the device chooses an audio signal associated with that species.

The response can include signals drawn from bioacoustics, the study of how animals produce, receive, and react to sound. Flox combines those signals with computer vision and behavioral observations. Its goal is to guide an animal away without physical contact.

This sequence distinguishes adaptive wildlife deterrence from a timer playing the same recording. A fox, goose, moose, and bear do not share the same sensory priorities. They also do not react identically when feeding, migrating, guarding territory, or moving with offspring.

Context creates a demanding technical problem. A camera must recognize animals at different distances and angles, sometimes in rain, snow, glare, darkness, or vegetation. It must also avoid mistaking people, shadows, vehicles, and moving plants for target wildlife.

The system then needs an appropriate response. An incorrect classification can trigger an irrelevant sound. A correct classification can still produce no movement if the animal is stressed, distracted, or already accustomed to similar signals.

Timing matters as well. A warning delivered before an animal enters a rail corridor has more value than one delivered after a train arrives. Flox advertises a response time below two seconds, but real-world safety depends on the entire detection and deployment geometry.

The technology’s strongest supporting evidence comes from specific operating trials. Between 2022 and 2024, Flox worked with Swedish airport operator Swedavia at Malmö, Kiruna, and Umeå airports.

Flox reports 100 percent deterrence across more than 200 animals in those trials. The tested wildlife reportedly included predatory birds, geese, foxes, and ungulates. The airport validation results come from the company, so they should be treated as vendor-reported findings.

Mora Airport offers a longer deployment example. Flox says magpies, crows, and other animals continued staying away from protected runway areas after six months. The airport also reported less dependence on manual patrols and improvised scaring.

These examples indicate that adaptive sound can work under selected conditions. They do not establish an independent performance rate across all airports, species, seasons, or landscapes. The scale and study design behind each claim remain essential context.

Rail testing introduces a different operating model. Flox and Alstom placed AI cameras on trains to observe wildlife beside several Swedish lines. The first phase detected animals, while a second phase added sound deterrence in April 2026.

The Swedish rail trials identified moose, roe deer, foxes, wild boar, farm animals, and birds. Alstom said the system performed particularly well when identifying farm animals, crows, and pigeons.

Moose and roe deer required more training to reach comparable identification accuracy. That admission is significant because it shows why field data matters. The species posing a serious collision risk may also be among the harder ones to recognize reliably.

Flox says its models learn from recorded encounters and its responses vary to discourage habituation. Yet “learning” can describe several different processes. It might involve periodic model updates, centrally designed playlists, or automated adaptation within each device.

Operators will need clarity about which decisions happen autonomously. They will also need controls governing updates, event review, false alarms, and intervention limits. Safety customers rarely purchase an algorithm without purchasing an accountability process around it.

The company’s longer-term plan extends the software into trains, cars, drones, tractors, and outdoor robots. That shift would move Flox from stationary deterrence toward a wildlife-perception layer embedded in moving machines.

Such a layer might warn a driver, adjust a machine’s route, or activate a deterrent. However, each action carries different safety requirements. Recognizing an animal is not equivalent to controlling a vehicle around one.

Adaptive Deterrence Still Has to Beat Established Alternatives

Flox is not competing against inaction. It must outperform a layered mix of barriers, people, habitat changes, sensors, sounds, dogs, and targeted removal.

The FAA recommends integrated wildlife management rather than reliance on one device. Airport programs can modify habitats, remove food or water sources, disperse animals, manage vegetation, improve fencing, and remove wildlife when necessary.

An adaptive sound system fits inside that framework. It does not eliminate the need for inspections, reporting, habitat work, or emergency response. A responsible deployment would use Flox as another layer, with its performance compared against existing site data.

Several companies already automate parts of this process. Robin Radar’s Scarecrow product identifies birds and plays species-specific distress calls. It also records activity so operators can review what happened.

iCHASE takes another route. Its system uses computer vision to detect birds and aims a moving laser toward them. The response changes with the target, giving operators an automated visual deterrent instead of an audio-centered one.

These competitors reveal the real primary opponent: adaptive systems versus static deterrence. Flox does not need to prove sound is universally superior to lasers, radar, dogs, or fencing. It needs to show that perception-linked responses stay useful longer than repeated, context-blind signals.

Its broader species ambition may help. Airport bird systems often focus on airborne threats, while Flox discusses animals ranging from geese to moose and grizzly bears. A multi-species platform could serve customers managing several kinds of infrastructure.

Breadth also creates risk. A model trained to distinguish birds around a Swedish airport may not transfer cleanly to mammals near an Australian road. Different regions bring new species, behaviors, legal constraints, climates, and acoustic environments.

False negatives are the most obvious safety concern. A missed animal leaves the original hazard in place. False positives create different costs, including unnecessary sounds, distracting alerts, disturbed wildlife, and declining trust among operators.

There is also a welfare question. “Humane” deterrence cannot be assumed simply because a device avoids lethal force. Sound can affect target animals, non-target species, nearby residents, and communication between animals.

Operators need evidence about sound intensity, exposure duration, response direction, and unintended displacement. Moving an animal away from one runway is useful only if the system does not push it toward another hazard.

Habituation remains another uncertainty. Flox says variable signals prevent animals from learning that the device poses no threat. Long-term, independently reviewed data would provide stronger evidence than short trials or customer testimonials.

The company’s airport claim of 100 percent deterrence across more than 200 animals sounds impressive. However, that figure does not reveal every denominator an infrastructure buyer needs.

Buyers should ask how a successful deterrence event was defined. They should also ask how many detections were missed, how long animals stayed away, and whether performance changed by species.

Weather and maintenance will influence results. Batteries lose capacity, lenses collect dirt, vegetation blocks views, and speakers degrade. A device installed in a controlled pilot may receive more attention than one among thousands of production units.

Cybersecurity and update procedures also matter when equipment enters critical infrastructure. Operators need to know who can change models or sound libraries. They need logs showing what the system detected, selected, and transmitted.

Local processing reduces dependence on network connectivity, but it does not remove software risk. Fleet dashboards, model distribution, compliance records, and customer accounts still create connected components requiring oversight.

Commercial execution may become the decisive constraint. Flox wants to expand from roughly 150 devices to several thousand while maintaining hardware, software, field support, and species-specific datasets. That is a substantial operational jump.

The company’s expected 2026 revenue remains modest compared with that ambition. Its proposed $10 million financing would support the expansion, but raising capital does not itself validate demand or field reliability.

Flox must demonstrate repeatable installation and support across customers, not only technical promise. Airports, railway operators, and public agencies often buy cautiously because equipment failures can carry safety and regulatory consequences.

The strongest version of the company’s case does not require every operator to abandon existing tools. It requires Flox to become the intelligent layer coordinating where, when, and how deterrence occurs.

What Will Show Whether Flox Can Scale

Three signals will determine whether Flox Intelligence wildlife AI becomes infrastructure or remains a promising collection of field trials.

The first signal is independent, species-level performance data. Future reports should separate detection accuracy from deterrence success. They should also show missed detections, false alarms, return times, seasonal variation, and results under difficult weather.

This evidence would strengthen Flox’s case if performance remains consistent across several independently operated sites. Large differences between company trials and customer-controlled evaluations would weaken it.

The second signal is the outcome of Alstom’s railway work. The partnership matters because it moves the software from stationary units toward equipment traveling through changing landscapes. Success requires fast recognition, suitable deterrence, and safe integration with railway operations.

Watch whether the partners publish collision changes or measurable reductions in high-risk wildlife presence. Detection counts alone show that animals are present. They do not show that the technology prevents accidents.

The third signal is deployment growth following the proposed Series A. Flox wants to move from about 150 devices to several thousand. Signed customers, repeat orders, renewals, and wider OEM adoption would provide stronger commercial evidence than the fundraising target.

Scaling should also reveal whether the subscription model works. Customers must see enough continuing value in monitoring, model updates, reporting, and analytics to renew after the initial hardware purchase.

The company’s broader vision deserves attention because it reaches beyond deterrence. Nozkova argues that machines operating outdoors need a better understanding of living environments. In that framing, wildlife recognition becomes another part of a machine’s perception stack.

That idea has applications across transport, agriculture, conservation, utilities, mining, and construction. A train could recognize animals before a collision. A tractor could avoid a nest, while a drone could alter its route around protected wildlife.

Still, each use case demands evidence and controls suited to its consequences. A farm deterrent, runway safety unit, and vehicle-control system cannot share one undifferentiated standard. The acceptable error rate changes with the possible harm.

For developers, the interesting lesson is that edge AI increasingly involves behavior, timing, and context rather than classification alone. A correct label has limited value if the system chooses the wrong action afterward.

For infrastructure buyers, the practical question is more direct. Does the system reduce incidents, patrol time, delays, and animal harm over several seasons? Dashboard activity and detection totals are supporting metrics, not final outcomes.

For conservation teams, the challenge is to measure where displaced animals go. An effective intervention should protect wildlife without simply moving conflict beyond the monitored boundary.

Flox Intelligence wildlife AI now has enough deployment history to warrant serious attention, but not enough public evidence to settle those questions. Its next phase will test whether adaptive deterrence works across species, climates, and customer operations.

The most useful action for prospective buyers is to demand outcome-based pilots with clear baselines. Track collisions, incursions, false alerts, repeat visits, maintenance, and non-target effects before and after installation.

If Flox publishes those results and converts trials into renewals, it will support a larger claim about outdoor AI. If the evidence remains limited to selected success stories, adaptive wildlife deterrence will remain an appealing mechanism without a proven infrastructure standard.