The Russian agency that interfered in the 2016 US election created a fake leftwing news publication, staffed it with fake editors with AI-generated photos and hired real freelance reporters as part of a fresh influence operation detected and removed by Facebook, the company said on Tuesday.
The latest operation by the Internet Research Agency (IRA) was still in its early stages when it was detected thanks to a tip from the FBI, according to Facebook’s head of security policy, Nathaniel Gleicher. The network had 13 accounts and two pages, with about 14,000 total followers.
The Facebook accounts and pages were designed to bolster PeaceData.net, an English- and Arabic-language website that claims to be a “global news organization”, but whose editorial staff are fictitious. Headshots of PeaceData’s “staff” were created using Generative Adversarial Networks, a type of AI that can produce lifelike images of faces, according to Graphika, a social media analysis firm that produced a report on the IRA operation.
“They put substantial effort into creating elaborate fictitious personas, trying to make fake accounts look as real as possible,” Gleicher said.
Many of the characters had profiles on Twitter and LinkedIn. Twitter said on Tuesday that it had suspended five accounts associated with PeaceData for “platform manipulation that we can reliably attribute to Russian state actors”. The company said tweets from the accounts were “low quality” and “spammy”, and that it would block links to content from PeaceData. LinkedIn did not immediately respond to a query.
Much of PeaceData’s content was copied from other websites, though some was produced by unwitting freelance reporters. Advertisements on Upwork and Guru.com offered a flat rate of $75 to entry-level writers. Major topics for the site included armed conflict, human rights abuses (especially by the US and UK), corruption, and the environment, as well as WikiLeaks, the coronavirus pandemic and the baseless QAnon conspiracy theory.
Four freelance journalists who wrote for PeaceData told the Guardian that they had been approached by one of PeaceData’s “editors” on Twitter, LinkedIn or by email with an offer to write for the site. Two were early-career writers who had recently been laid off and were eager to establish themselves; two were more experienced writers. The Guardian agreed to let them speak anonymously because they were concerned for their careers.
The writers only learned of the deception from news reports or reporters’ inquiries. One of the experienced journalists said that PeaceData had paid $250 up front, which was unusual, then “ghosted” her after publishing one piece. “I didn’t imagine a scam would have paid me up front like that,” she said.
Another writer said he was approached via direct message on Twitter and offered $200-$250 a piece, more than he was usually paid for writing.
“I was just trying to get more bylines and get paid to do what I want to do,” he said. “I’ve interacted with editors who do far less than what they were doing, and they paid faster than some publications ... I’m a freelance writer – I’m used to being taken advantage of.”
PeaceData’s coverage of the US portrayed the country as “war-mongering and law-breaking abroad while being racked by racism, Covid-19, and cutthroat capitalism at home”, according to the report. The outlet was negative toward Donald Trump, but Graphika found that its treatment of his Democratic rival Joe Biden and vice-presidential nominee Kamala Harris was “noteworthy for its hostile tone”.
The US-focused content of PeaceData appeared designed to “build a leftwing audience and steer it away from Biden’s campaign”, according to the Graphika analysis. UK-focused content similarly appeared to appeal to leftwing audiences with attacks on the Labour party leader, Keir Starmer, for being too centrist.
The operation targeted supporters of Bernie Sanders and democratic socialists in the US and supporters of Jeremy Corbyn in the UK by having one of the fake accounts, the fictitious “Alex Lacusta”, post links to PeaceData articles in affiliated Facebook groups.
The IRA also used “unwitting users” to attempt to obtain authorization from Facebook to run political ads in the US, the company said. Facebook implemented the authorization process for political advertisers after the 2016 election, when the IRA was able to spend about $100,000 – some of it in rubles – on ads that targeted US voters with divisive messaging.
An unsigned statement posted on PeaceData’s website said allegations the site was “a Russian propaganda tool” were “an ugly lie”.
Both Facebook and Graphika concluded that the operation had been detected and taken down before causing significant damage.
“It follows a steady pattern where particularly Russian actors have gotten better at hiding who they are, but their impact is smaller and smaller and they are getting caught earlier,” Gleicher said. “These actors are caught between a rock and hard place: run a large network that gets caught quickly or run a small network that has limited reach.”
“The operation’s greatest success – to the extent that it had any – lay in its ability to co-opt unwitting authors to write its content,” the Graphika analysis concludes. “The IRA’s 13 accounts managed to deceive that pinpoint audience; they do not appear to have reached a substantially larger one.”
Twitter appealed to “governments around the world” to stop attempting to deceive users through similar operations.
“Regardless of the low-level impact in this case, governments around the world must stop these practices,” the company said in a tweeted statement. “They’re antidemocratic. Attempts to manipulate our service to undermine democracy – by both foreign and domestic actors – will be met with strict enforcement of our policies.”
Agencies contributed reporting