The Washington PostDemocracy Dies in Darkness

Facebook, Twitter disable sprawling inauthentic operation that used AI to make fake faces

The Facebook logo displayed on a tablet in Lille. (Denis Charlet/AFP/Getty Images)

Facebook and Twitter on Friday disabled a global network of hundreds of fake accounts that pushed pro-Trump messages and covered its tracks using inauthentic photos generated with the aid of artificial intelligence.

The use of AI to trick social media companies, deceive unsuspecting users and essentially create people who do not exist marked a major, troubling new development, according to disinformation researchers, who expressed fresh alarm that such tactics could have implications for the 2020 presidential election.

The tech giants’ takedown targeted the BL, a U.S.-based media company that Facebook linked to Epoch Media Group. The organization has criticized the Chinese government, and it has ties to Falun Gong, a Chinese spiritual movement. It also has vociferously supported the reelection of President Trump.

The researchers with whom Facebook shared data about the BL and Epoch Media Group described the operation in a report as “a large-scale artificial amplification factory."

Facebook CEO Mark Zuckerberg says in interview he fears ‘erosion of truth’

Facebook in particular said the BL was linked to hundreds of fake accounts spread across its services, which posted political messages at high frequencies, often in an attempt to direct traffic back to their websites. The social media companies signaled they did not take action because of the content of those posts but, rather, the tactics deployed by those who engaged in them, such as the use of AI-generated images, which violated rules prohibiting spam, misrepresentation and coordinated inauthentic behavior.

Disinformation experts at Graphika and the Atlantic Council’s Digital Forensic Research Lab — which were given access to Facebook data in advance of the takedown announced Friday — said it marked the “first time” they had seen “AI-generated pictures deployed at scale to generate a mass collection of fake profile pictures deployed in a social media campaign.”

“The BL is now banned from Facebook,” said Nathaniel Gleicher, the head of security policy at Facebook. “We are continuing to investigate all linked networks, and will take action as appropriate if we determine they are engaged in deceptive behavior.”

On Facebook, the network it disabled encompassed more than 600 accounts, with dozens of pages and groups that purchased $9 million in ads. Roughly 55 million users on Facebook followed at least one of the pages tied to the operation, though the company did not specify how many of those people were based in the United States.

Facebook takedowns show new Russian activity targeted Biden, praised Trump

Twitter, for its part, confirmed it had suspended 700 accounts linked to that same network for violating “rules around platform manipulation — specifically fake accounts and spam.” Google did not respond to a request for comment.

In response, Epoch Media Group’s English-language edition publisher, Stephen Gregory, said in a statement the organization “has no connection with the website BL." Rather, he said the BL employs its former workers, and it is a publication of Epoch Times Vietnam, which “was no longer listed as part of Epoch Media Group” as of October 2018.

“Epoch Media Group calls upon Facebook to withdraw its claim that Epoch Media Group has been involved in coordinated action with BL and to withdraw its claim that BL has been working on behalf of Epoch Media Group,” Gregory said.

The removals announced Friday illustrate the fast-evolving, increasingly complicated task social media giants face in battling back viral falsehoods, fake accounts and other troublesome content on the Web. More malicious actors now seek to manipulate conversation online using new, sophisticated techniques just four years after Russian agents weaponized Facebook, Google and Twitter to undermine the 2016 presidential race.

To that end, Facebook and Twitter announced a slew of additional enforcement actions on Friday targeting coordinated, inauthentic behavior originating out of countries such as Georgia and Saudi Arabia. “We are making progress rooting out this abuse,” Facebook’s Gleicher said. “But as we’ve said before, it’s an ongoing challenge."

News outlets and researchers had raised alarms about the BL for months, pointing to its suspicious tactics and the possibility it had created fake accounts to amplify pro-Trump messages. Snopes, in particular, first linked the organization with the Trump-leaning Epoch Media Group, though officials there at the time denied it. Facebook on Friday affirmed the connection.

Pages tied to the BL posted a wide array of content, experts said, including clickbait about animals and politically hostile posts about the Chinese government. But English-language pages associated with the organization “focused heavily on positive coverage” of Trump, while attacking his critics, as well as the researchers at Graphika and DFR Lab. Fake accounts ran many of the pages and initiated many of the engagements with the posts, photos and other content.

Some of those fake accounts relied on AI tools and technology to create pictures of people who do not exist — fake photos that, while believable, still contained quirks that helped Facebook and its researchers spot them. Others blatantly stole photos from real users, researchers said. And the fake accounts may have relied on publicly available technology to publish their posts, tweets and other content in batches, amplifying their reach.

Before imposing the ban Friday, Facebook had started taking action against the network’s advertising: Over the summer, for example, it removed an ad that included a video of Trump decrying “fake news” and asking users to text “MAGA” to a listed number to get “FREE real news.” But limiting the group’s advertising did little to arrest the circulation of its content on social media.

The “About us” page for the BL — short for “The Beauty of Life” — offered insight into how the site presented itself to Facebook users. It said the group’s mission was to “present to the world the most beautiful aspects of life.” And it presented itself as a bulwark of trust in a moment of suspicion about information and how it is consumed.

“Inaccurate and degenerate information can be easily channeled toward vulnerable or uninformed people,” its since-disabled page read, “creating a vicious cycle of misinformation.”