Instagram blocked the search terms. Its algorithm serves the videos anyway

The genre is simple enough to describe. A man wearing camera glasses walks up to a woman in the street. He delivers a line and films her reaction without asking. The clip goes to Instagram, tagged as shot on Meta glasses. The woman is now content. Some of these videos have millions of views.

Adam Mosseri, who runs Instagram, said in mid-July that this had to stop. Meta would remove “harassing” pickup videos under its existing rules on bullying and harassment. We covered that promise when he made it.

“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 at the time.

Nine out of more than 20

Business Insider spent the following month looking. Katie Notopoulos found dozens of the videos still live, several from popular and verified accounts. She sent Meta a list of more than 20. The company took down nine.

One removal is instructive. A creator had used the same line on several different women across several videos. Meta deleted the one on the list. The others stayed up.

Meta does not dispute that it is acting. A spokesperson, Tracy Clayton, said thousands of pieces of content have been removed and several large accounts taken down. Instagram now blocks searches for “rizz” and “cold approach”, the slang terms for this content. Those are real enforcement actions, and no platform moderates perfectly at this scale.

The pattern around them is the problem. One large account deactivated in the first wave came back a few days later. Meta said that was an error and re-banned it once Business Insider pointed it out. Meta deactivated a second account, with 174,000 followers, the morning after it appeared on the reporter’s list. Its owner appealed. A day later it was back, videos restored.

This is not only an Instagram problem. TikTok took down two of the four videos it received, citing bullying and sexual harassment. YouTube removed several videos and pulled one channel from its monetisation programme.

The feed does the finding now

Here is the part the follow-up coverage skipped. Notopoulos reports that after she sought out and interacted with these videos, Reels began serving her more of them, until her feed was filled almost entirely with the genre. Search terms stopped being necessary.

Meta has blocked the words people use to find this content. The recommendation system then finds it for them. Both systems belong to the same company, and they are pulling in opposite directions.

That matters more than any individual takedown. Blocking a search term stops deliberate hunting. It does nothing about a ranking model optimised for whatever holds attention, and this content evidently holds it. Moderation removes items. Recommendation decides how many people see the ones that remain.

Researchers have started counting

The claim is no longer only journalistic. Four researchers at the University of Sydney, Joanne Gray, Milica Stilinovic, Marcus Carter and Ben Egliston, watched hundreds of these videos and published a preprint last week. The title is “Harm Through the ‘#RizzCam’: Smart Glasses, Ambient Capture and Invisible Harassment”.

404 Media reported the paper on 12 August. The work is a preprint, so it has not yet been peer reviewed.

Carolina Are, a digital criminologist at the London School of Economics, told Business Insider that Meta’s response misses the level the problem sits at.

“There is a backlash, and then Meta tends to minimize its own responsibility for it. They talk about how this is user-generated behavior or an error in enforcement,” she said. “But they don’t recognize the systemic issues that are causing that to happen.”

A problem Meta built and sells

The reason Mosseri personally delivered this message, when his counterparts at other platforms did not, is that Meta is on both sides of it. The glasses are its product. The feed is its product. The harassment sits between them.

The hardware is working commercially. Sales topped 7 million units in 2025, more than triple the year before, and Counterpoint Research puts Meta clearly ahead in smart glasses. That makes them a rare success inside Reality Labs, which is still losing money and cut jobs this year. Mark Zuckerberg has argued that people without AI glasses will end up at a cognitive disadvantage to those who have them.

Reputation is the exposure, and the backlash has been building for two months. A $2 sticker defeated the recording light. The glasses were banned from UK pubs and then from courts in England and Wales. Commenters filled Kylie Jenner’s posts calling her Meta eyewear “pervert glasses”.

Automating the cleanup

Meta’s answer to moderation at scale is more automation. It told the Financial Times in June that early tests showed its AI systems flagged more violating content than human moderators while making fewer mistakes. Whether that holds at scale is unproven.

Sarah T. Roberts, who co-founded the UCLA Center for Critical Internet Inquiry, put the timeline plainly to Business Insider.

“It’s not like they just started those platforms yesterday,” she said. “It’s not like they don’t have 20 years now of understanding how people will behave at the most depraved levels if given a chance.”

The enforcement will probably improve. The harder question is the one nobody at Meta has answered. The company sells a camera that is designed not to be noticed, and it runs the feed that rewards what the camera captures. Removing the videos one at a time treats the output. It leaves the machine that produces them running.