Your Phone Thinks You’re a Thief. That’s Not a Bug—It’s a Feature.

You laced up your running shoes, stepped outside, and started jogging. Two minutes later, your phone screamed an alert: “Someone is snatching your phone and running off.” Congratulations—you’ve just been flagged as a criminal for going for a run.

This isn’t a glitch. It’s not a misconfiguration. It’s a product philosophy made visible. And it’s happening to thousands of people right now.

The algorithm doesn’t see a person exercising. It sees a thief in motion. That’s the difference between a paranoid threat model and actual human behavior.

Google’s theft detection feature, launched in Android 15, uses accelerometer and motion sensors to detect a sudden run—the kind of motion a thief would make after snatching a phone. The problem? Running is also what people do when they’re late for a bus, playing with their kids, or, you know, exercising. In one tester’s account, the feature triggered so often that the user gave up and turned it off. As one commenter quipped, “After looking over Google’s advice, my key takeaway was that thieves are in much better shape than me, and run both longer and faster than this nerdy writer has in more than two decades.”

You’ve probably felt that sting of absurdity when technology punishes you for being normal. The phone that locks you out because you walked too fast. The app that asks if you’re a robot because you scrolled too smoothly. Each false positive isn’t a bug—it’s a trade-off that designers made invisible. They optimized for the rare, terrifying event (theft) and ignored the common, boring event (running). And in doing so, they encoded a paranoid worldview where sudden movement equals danger, and everyday life becomes the anomaly.

Every false positive is a choice. Every alert is a statement about who the system trusts—and who it doesn’t.

This isn’t just about Google. It’s about every safety feature that trades usability for vigilance. The car alarm that goes off when a cat walks by. The fraud detection that blocks your credit card because you bought gas in a different state. The surveillance camera that flags a mail carrier as a trespasser. We’ve been trained to accept these inconveniences as the price of safety. But the price is higher than we think. Each false alarm erodes trust, conditions us to ignore warnings, and ultimately makes us less safe when a real threat appears.

I saw this firsthand when a friend’s phone locked them out of their own car because the motion sensor thought they were being carjacked. They were just reaching for their coffee. The algorithm didn’t know the difference. It never will, because it was never designed to.

Here’s the twist: the feature isn’t a failure. It’s working exactly as intended. The engineers optimized for a specific threat—a snatch-and-run—and they succeeded. The problem is that the threat model is a caricature, not a reality. Real life is messy, full of ambiguous motions that don’t fit into neat categories. When you design for the worst-case scenario, you end up punishing every scenario that looks like it.

The safest product in the world is useless if it makes normal life feel like a crime scene.

This matters to you because you’re already living in a world of invisible trade-offs. Every safety feature you use—from biometric locks to fraud alerts—is a bet on what’s normal and what’s dangerous. And the people making those bets are usually not you. They’re engineers in a lab, testing against idealized scenarios, not against your morning jog. The question isn’t whether the feature works. It’s whether the feature works for you, in your actual life.

So next time your phone accuses you of theft, take a moment to ask: Who decided that running is suspicious? Who decided that your normal is a threat? The answer is uncomfortable. It’s the same people who decided that convenience is less important than caution. And until we start demanding that our safety features respect the reality of our lives, we’ll keep getting flagged for being human.