Flox's edge AI is scaring bears off train tracks

Wildlife and heavy infrastructure have always been a brutally expensive mix.

In the U.S., wildlife strikes cost civil aviation about $272 million every year. In France, roaming animals disrupt more than 8,000 trains annually and rank among SNCF's top five causes of rail delays. Traditional fixes have barely changed in decades: put up taller fences, blast sirens, flash strobe lights, or cull herds. The catch is that wild animals get used to fixed sirens surprisingly fast.

Stockholm-based startup Flox Intelligence is tackling the problem with on-device AI. Spun out of the KTH Royal Institute of Technology in 2020 by transportation veteran Sara Nozkova, the company builds battery-powered edge devices designed to scare animals off without harming them.

Smarter deterrents, not louder sirens

Instead of blasting noise around the clock, Flox boxes stay dormant until a motion sensor trips. The device wakes up its camera, runs an on-device computer vision model to identify whether that moving shape is a moose, deer, or bear, and then plays a species-tailored sound to steer it away.

Crucially, the system varies its acoustic signals based on how the animal reacts, preventing wildlife from getting habituated to the noise over time.

Flox sells the hardware upfront, followed by an annual software subscription covering monitoring, analytics, and regulatory compliance reports. It is also licensing software directly to vehicle manufacturers, including train maker Alstom.

Around 150 Flox units are currently operating across North America, Europe, and Australia along rail lines, airport boundaries, mining facilities, and electrical substations. The startup expects around $1.5 million in revenue for 2026 and is now opening a $10 million Series A round to scale deployments into the thousands.

Why it matters

For transit agencies and airfield managers, Flox turns an unpredictable environmental liability into a measurable software line item. If varying predator calls and biological signals can keep a runway clear or stop a high-speed train from hitting a 1,000-pound animal, the payback period is immediate compared to train repairs and delay fines.

Across the broader AI landscape, this marks an interesting push into unscripted physical environments. Autonomous driving companies spent billions mastering predictable city grids, but nature perception remains largely unmapped. While competitors like Robin Radar and iCHASE tackle bird strikes using distress calls or moving lasers, Flox is trying to map large mammal behavioral data in the wild.

We think Flox's go-to-market strategy is clever: bundling edge hardware with OEM vehicle integrations gives it multiple distribution paths without having to lay miles of physical cabling. The big test ahead is proving that wild animals truly cannot outsmart the algorithm over years of exposure. At just 150 deployed units, Flox still needs to demonstrate that its dynamic sounds work consistently across diverse geographies before claiming it has solved habituation for good.