Meta promised to crack down on creeps using its AI glasses to secretly film women on the street — and weeks later, the videos are still all over Instagram, which means any American walking down the sidewalk can become unwilling content for some stranger's viral reel.

In mid-July, Instagram head Adam Mosseri said the platform would remove "harassing" pickup videos filmed with Meta's smart glasses. "We don't want people to be surreptitiously taking videos of other people and harassing them and then posting them on our platform," Mosseri said. But a Business Insider investigation found dozens of these videos still live on the platform weeks later, some from verified creators with large followings. One video shows a man in Meta glasses asking a woman, "Are you a secret code?" When she looks confused, he follows up: "Because I'm trying to crack you." The glasses record the whole thing without her consent.

When Business Insider shared more than 20 flagged videos with Meta, the company removed just nine. One deactivated account was briefly restored — a Meta spokesperson called it an error — and only re-banned after BI pointed it out. Meta spokesperson Tracy Clayton says thousands of pieces of content have been removed and search terms like "rizz" and "cold approach" are now blocked. That sounds aggressive. In practice, the enforcement is Swiss cheese.

Here's the real problem: Instagram's algorithm is built to feed you more of whatever keeps you scrolling. According to Digital Trends, interacting with just a few of these videos during the investigation was enough for Reels to start flooding the user with similar content. Meta's own discovery engine is doing the work its moderation team can't — serving up secretly filmed women to anyone who lingers on a single clip.

This isn't just a moderation failure. It's a business model laid bare. Meta has sold more than 7 million pairs of smart glasses in 2025, triple the prior year, per Digital Trends. These glasses are central to Mark Zuckerberg's push into AI-powered wearable computing. Some people already call them "creep glasses." The more that label sticks, the more it threatens the hardware play.

Carolina Are, a digital criminologist at the London School of Economics, told Business Insider that Meta "minimizes its own responsibility" and "takes no accountability for what their policies or infrastructure have done to enable that behavior." She's right. You build a camera that sits on someone's face, you build an algorithm that amplifies whatever gets engagement, and then you act surprised when the combination produces exactly what any sane person predicted.

Meta says AI will eventually fix its moderation at scale, claiming early tests show machines catching more violations with fewer errors than humans. Maybe. But the company has also built a machine that profits from distributing the violations in the first place.

The question isn't whether Meta can train an AI to catch what its other AI promotes. The question is why anyone should trust a company that needed a reporter's phone call to re-ban an account it had already banned once.