Flock Cameras: When AI Surveillance Creates a Crash Risk

There's a Flock camera bolted to a pole near the on-ramp I take every morning. Last week I watched the driver ahead of me stomp the brakes so hard their rear bumper dipped—just because they wanted to gawk at the blinking license plate reader. That instinct to look is exactly why a growing number of US traffic safety experts are now saying these surveillance cameras can become a crash risk themselves.

The concern isn't that the cameras are spying on you. It's that they're deployed like traffic furniture without enough thought about human behavior. Drivers see a device they don't recognize, or one that emits a sudden IR flash at night, and they react the way humans always react: they slow down, swerve, or rubberneck. A few seconds of hesitation on a 45 mph road is enough to cause a rear-end collision. And in a real-world scenario, that's a far more immediate danger than the vague privacy debates we usually get stuck on.

Red-light cameras already taught us this lesson years ago. Jurisdictions that hung them too close to intersections saw higher rear-end crash rates because drivers braked abruptly to avoid tickets. Flock's cameras are marketed mainly for automated license plate recognition (ALPR), so they're often placed on utility poles right alongside active lanes, sometimes with no setback and no warning signage. Instead of treating the deployment like a safety engineering problem, many cities treat it like a simple IT install. That's a classic blind spot in how we roll out AI-powered infrastructure: we focus on the algorithm's accuracy, not on the physical context where the algorithm lives.

I'm not anti-surveillance-camera. The technology has genuinely useful applications—recovering stolen cars, finding missing people, and giving law enforcement a time-stamped trail. But a police department's operational convenience doesn't trump the basic traffic-engineering principle that anything in the right-of-way must not create an unnecessary distraction. Experts recommend simple fixes: place cameras at least a few feet behind the curb line, angle them so the flash isn't aimed at oncoming windshields, and add clear "photo enforcement" style signage so drivers are conditioned to ignore them. Those are cheap, obvious changes. The fact that they're not already standard practice tells you how little oversight there is over this hardware.

What bugs me is that this won't be the last time an AI deployment has this kind of predictable, physical side effect. If we can't even properly plant a camera on a pole, how are we going to handle autonomous vehicles or delivery drones sharing space with people? We need to stop treating every new sensor as a purely digital object. It's not. It's a roadside object with a lens, and it demands the same scrutiny we give to guardrails and traffic lights. Otherwise, we're just counting crashes after the fact instead of preventing them at the design stage.