“Watch your step sir, keep moving,” a police officer with a vest that reads ICE and a patch that reads “POICE” says to a Latino-appearing man wearing a Walmart employee vest. He leads him toward a bus that reads “IMMIGRATION AND CERS.” Next to him, one of his colleagues begins walking unnaturally sideways, one leg impossibly darting through another as he heads to the back of a line of other Latino Walmart employees who are apparently being detained by ICE. Two American flag emojis are superimposed on the video, as is the text “Deportation.”
The video has 4 million views, 16,600 likes, 1,900 comments, and 2,200 shares on Facebook. It is, obviously, AI generated.
Some of the comments seem to understand this: “Why is he walking like that?” one says. “AI the guys foot goes through his leg,” another says. Many of the comments clearly do not: “Oh, you’ll find lots of them at Walmart,” another top comment reads. “Walmart doesn’t do paperwork before they hire you?” another says. “They removing zombies from Walmart before Halloween?”
The latest trend in Facebook’s ever downward spiral down the AI slop toilet are AI deportation videos. These are posted by an account called “USA Journey 897” and have the general vibe of actual propaganda videos posted by ICE and the Department of Homeland Security’s social media accounts. Many of the AI videos focus on workplace deportations, but some are similar to horrifying, real videos we have seen from ICE raids in Chicago and Los Angeles. The account was initially flagged to 404 Media by Chad Loder, an independent researcher.
“PLEASE THAT’S MY BABY,” a dark-skinned woman screams while being restrained by an ICE officer in another video. “Ma’am stop resisting, keep moving,” an officer says back. The camera switches to an image of the baby: “YOU CAN’T TAKE ME FROM HER, PLEASE SHE’S RIGHT THERE. DON’T DO THIS, SHE’S JUST A BABY. I LOVE YOU, MAMA LOVES YOU,” the woman says. The video switches to a scene of the woman in the back of an ICE van. The video has 1,400 likes and 407 comments, which include “ Don’t separate them….take them ALL!,” “Take the baby too,” and “I think the days of use those child anchors are about over with.”
The USA Journey 897 account publishes multiple of these videos a day. Most of its videos have at least hundreds of thousands of views, according to Facebook’s own metrics, and many of them have millions or double-digit millions of views. Earlier this year, the account largely posted a mix of real but stolen videos of police interactions with people (such as Luigi Mangione’s perp walk) and absurd AI-generated videos such as jacked men carrying whales or riding tigers.
The account started experimenting with extremely crude AI-generated deportation videos in February, which included videos of immigrants handcuffed on the tarmac outside of deportation planes where their arms randomly detached from their body or where people suddenly disappeared or vanished through stairs, for example. Recent videos are far more realistic. None of the videos have an AI watermark on them, but the type and style of video changed dramatically starting with videos posted on October 1, which is the day after OpenAI’s Sora 2 was released; around that time is when the account started posting videos featuring identifiable stores and restaurants, which have become a common trope in Sora 2 videos.
A YouTube page linked from the Facebook account shows a real video uploaded of a car in Cyprus nearly two years ago before any other content was uploaded, suggesting that the person behind the account may live in Cyprus (though the account banner on Facebook includes both a U.S. and Indian flag). This YouTube account also reveals several other accounts being used by the person. Earlier this year, the YouTube account was posting side hustle tips about how to DoorDash, AI-generated videos of singing competitions in Greek, AI-generated podcasts about the WNBA, and AI-generated videos about “Billy Joyel’s health.” A related YouTube account called Sea Life 897 exclusively features AI-generated history videos about sea journeys, which links to an Instagram account full of AI-generated boats exploding and a Facebook account that has rebranded from being about AI-generated “Sea Life” to an account now called “Viral Video’s Europe” that is full of stolen images of women with gigantic breasts and creep shots of women athletes.
My point here is that the person behind this account does not seem to actually have any sort of vested interest in the United States or in immigration. But they are nonetheless spamming horrific, dehumanizing videos of immigration enforcement because the Facebook algorithm is rewarding them for that type of content, and because Facebook directly makes payments for it. As we have seen with other types of topical AI-generated content on Facebook, like videos about Palestinian suffering in Gaza or natural disasters around the world, many people simply do not care if the videos are real. And the existence of these types of videos serves to inoculate people from the actual horrors that ICE is carrying out. It gives people the chance to claim that any video is AI generated, and serves to generally litter social media with garbage, making real videos and real information harder to find.
Meta did not immediately respond to a request for comment about whether the account violates its content standards, but the company has seemingly staked its present and future on allowing bizarre and often horrifying AI-generated content to proliferate on the platform. AI-generated content about immigrants is not new; in the leadup to last year’s presidential debate, Donald Trump and his allies began sharing AI-generated content about Haitian immigrants who Trump baselessly claimed were eating dogs and cats in Ohio.
In January, immediately before Trump was inaugurated, Meta changed its content moderation rules to explicitly allow for the dehumanization of immigrants because it argued that its previous policies banning this were “out of touch with mainstream discourse.” Phrases and content that are now explicitly allowed on Meta platforms include “Immigrants are grubby, filthy pieces of shit,” “Mexican immigrants are trash!” and “Migrants are no better than vomit,” according to documents obtained and published by The Intercept. After those changes were announced, content moderation experts told us that Meta was “opening up their platform to accept harmful rhetoric and mod public opinion into accepting the Trump administration’s plans to deport and separate families.”
“It’s expensive to take somebody else’s man”: Influencer ordered to pay $1.75 million for breaking up a marriage