Meta removed AI abuse videos only after journalists intervened

What users reported

The videos showed adult men touching women and girls in gymnastics leotards at gyms and dance studios. Some figures appeared to be teenagers, though their exact ages were not given. In near-photorealistic clips, coaches touched athletes’ lower backs and inner thighs, groped them, kissed them, or touched their genitals over clothing. The athletes often smiled or giggled, as if the contact were normal.

Kim Shore, a gymnast and safe-sport advocate who consults for Gymnasts for Change International, reviewed many of the pages. She told Futurism that the videos depicted abuse, and warned that they could make coercive contact look like playful or helpful coaching in sports already marked by a power imbalance.

The clips were not all reaching tiny audiences. Some drew more than 100,000 views. One video recommended by Facebook had more than 176,000 views just two days after it was posted.

In one case, journalists reported ten pages from an ordinary Facebook account, selecting the categories “adult content” and “sexual exploitation.” A week later, Facebook said eight pages did not violate its rules and would remain online. Two had disappeared, though it was unclear whether the reports caused their removal.

After journalists sent Meta links to the pages and asked why the videos were allowed, all of them disappeared. Facebook then replaced several earlier support notices with messages saying the pages had been removed for violating its Community Standards.

Meta told Futurism that content depicting abuse or exploitation should not be on its platforms, whether realistic or AI-generated, and said it had removed the pages for violating its rules.

The rules were not the missing piece

Meta’s policies prohibit depictions of “impending sexual activity,” including poses suggesting that someone’s hand is about to touch another person’s genitals or anus. Its adult sexual exploitation policy covers people depicted nude or nearly nude, engaged in sexual activity, or in sexually suggestive poses—and explicitly includes digital images, including AI-generated ones.

Its policy on child sexual exploitation, abuse and nudity also prohibits sexualizing real or fictional children across photos, video, art, digital content and written descriptions. It bars genital or anal stimulation, including when it is only about to happen or is not directly visible.

That makes the initial decisions hard to explain as a gray area in the written rules. After the press inquiry, Meta acknowledged that it sometimes misses content, while saying it had proactively found and removed 98% of posts related to child endangerment in the previous quarter before anyone reported them.

The recommendations tell a less reassuring story. After the account filed reports, Facebook began serving it many similar AI videos of coaches touching, kissing and groping young athletes. The reporting process had not just failed to remove the material; the platform also appeared to use the account’s interaction with it to recommend more.

The harm is not confined to the screen

Shore’s concern is that the videos can teach young athletes to tolerate contact that should alarm them. Children in sports are often expected to trust coaches, follow instructions and accept physical contact as part of training. Shore said that combination can make it harder for a child to recognize or resist grooming and abuse.

She also warned that the clips could give adults who abuse their authority examples of how to present sexualized conduct as ordinary coaching, affection or play. That risk is her assessment, not a demonstrated outcome—but the videos are being distributed in a setting where, according to the 2021 World Players Association global study cited in the Russian report, 13% of child athletes experience sexual abuse in sport.

My read is that the most consequential failure here is not simply that moderators missed a set of pages. Meta’s own policies already describe the kinds of imagery at issue, yet ordinary users received a different result from journalists who raised the same material with the company’s press office. And the recommendation system then amplified similar videos to the reporting account.

Meta has said it is fighting abusive material and that it sometimes misses things. The unresolved tension is whether its safeguards work for the people who encounter this content first—or mainly when a failure becomes visible to the press.