Data Versus Democracy
Page 15
least enthusiasm—among those same potential Clinton supporters on her
left and in the middle. As Jeff Stein wrote for Vox:
None of the Podesta emails has so far actually broken any fresh
scandals about the woman on track to be the next president. Instead,
they’ve mostly revealed an underbelly of ugliness to the multiple
Clinton controversies that we’ve already known about: the ques-
tionable relationship between the Clinton Foundation and its
donors, Clinton’s ease with powerful interests on Wall Street, her
ties to wealthy campaign contributors. 31
The Clinton Foundation controversies were likely particularly damaging to
Clinton’s potential support among centrists turned off by Trump. Many of
them had voted for Republicans in the past and were dissatisfied by the GOP’s
nominee. But narratives of corruption around her, combined with Republican
prejudices against the Clinton name, likely made it difficult for a number of
moderates and dejected Republicans to vote Democrat in 2016. The ties to
Wall Street and wealthy donors likely hurt Clinton’s chances with former
Sanders supporters, who had been calling for Clinton to release the content
of the private speeches she had been paid to give to bankers and other wealthy
Americans in the past. Some of that content was awkward or embarrassing,
but some of it also undercut Clinton’s ability to garner support among those
to the left of her campaign’s policy platform.
30Ibid.
31Jeff Stein, “What 20,000 pages of hacked WikiLeaks emails teach us about Hillary
Clinton,” Vox, published October 20, 2016, www.vox.com/policy-and-politics/2016/
10/20/13308108/wikileaks-podesta-hillary-clinton.
82
Chapter 5 | Democracy Hacked, Part 1
I also think it is likely that the months-long obsession with “Clinton’s emails”
compounded the impact of FBI Director James Comey’s announcement on
October 28, 2016, that he was reopening the investigation into Clinton’s use
of a private email server while Secretary of State. That investigation was
reopened due to evidence that surfaced during the investigation of former
Congressman Anthony Weiner, whose estranged wife, Huma Abedin, was a
top Clinton aide.32 Comey was investigating the possibility that Clinton’s
private email server may have contained classified information, not the content
of emails published by Russia or WikiLeaks, but a fast-moving news cycle
replete with vague references to “Clinton’s emails” likely meant that these
news stories compounded each other’s impact on voters.
Perhaps most significantly, though, is that the GRU’s release of kompromat on
Clinton and the DNC meant that Russia got to set a significant portion of the
agenda of the presidential election. Journalists pored over leaked content looking
for scoops, which they reported. Those same journalists posed questions
about the content of these documents in debates. And the candidates faced
questions about these documents all along the campaign trail, from both
journalists and voters. Every question about this Russian kompromat—
however legitimate a concern that question represented—was a question
that didn’t address the policy platforms set forth by the campaigns. The hours
spent chasing down security concerns and prepping answers to these questions
were hours not spent by the campaigns building and defending their own
platform. To a nontrivial extent, Russian hackers determined the issues that
American voters would devote their limited attention to, and on which they
would determine how to cast their votes on November 8. My colleagues and
I call this manipulation of public messaging “weaponized truth.” The contents
of Fancy Bear’s data dumps appeared authentic to security researchers—they
were not “fake news.” But they were meant to manipulate public discourse
and individuals’ behavior. Whether or not we call that “disinformation” (I do,
because disingenuous manipulation is at the core of the operation), it is
certainly a component of information warfare—and because of its basis in
facts, it can be an incredibly effective one.
Project Lakhta
Of course, Russia did not just manipulate American public discourse via the
release of hacked and leaked documents. Their Internet Research Agency
(IRA) also conducted information warfare against the United States through
an expert social media manipulation campaign. The details of the IRA’s
extensive operations are detailed primarily in indictments issued in 2018 by
32Adam Goldman and Alan Rappeport, “Emails in Anthony Weiner Inquiry Jolt Hillary
Clinton’s Campaign,” The New York Times, published October 28, 2016, www.nytimes.
com/2016/10/29/us/politics/fbi-hillary-clinton-email.html.
Data versus Democracy
83
the U.S. Department of Justice against key Russian individuals and LLCs and in
two reports prepared for the U.S. Senate Select Committee on Intelligence
(SSCI) and released in December 2018, providing a detailed analysis of data
provided to SSCI by Twitter, Facebook (including Instagram), and Google
(YouTube). 33
The picture portrayed in these documents—which is likely still incomplete—
is one of large-scale, expert manipulation of public attention through a
combination of “weaponized truth,” partial truths, flat-out lies, and voter
suppression narratives, aimed at the election of Donald Trump as president of
the United States and discrediting the potential presidency of Hillary Clinton
and the democratic process in general. Further, after the election, the IRA
continued to attempt to manipulate and destabilize American society, even
increasing their activity aimed at certain American communities on platforms
like Instagram. And though the data currently available suggests that
government and platform activities have significantly hindered the IRA’s ability
to wage social media–based information warfare, it is also clear that Russian
groups are still attempting to manipulate public opinion and discredit their
critics via U.S.-targeted online media.
While Russian influence operations go back much further, even in the United
States, the now-famous operation aimed at manipulating the 2016 election
ratchets up in April 2014 with the formation of the IRA’s “translator project,”
aimed at studying U.S. social and political groups on online platforms, including
YouTube, Facebook, Instagram, and Twitter. In May 2014, the strategy for this
operation, known internally as “Project Lakhta,” was set: “spreading distrust
towards the candidates and the political system in general”34 with the objective
of interfering in the 2016 U.S. election, specifically. 35 By June 2014, IRA
operatives were already conducting in-person intelligence gathering physically
in the United States (with a subsequent trip in November 2014).36 By
September 2016, Project Lakhta’s monthly budget was approximately 1.25
million dollars. 37
According to the IRA indictment:
Defendants and their co-conspirators, through fraud and deceit,
created hundreds of social med
ia accounts and used them to develop
certain fictitious U.S. personas into “leader[s] of public opinion” in
the United States.
33Full disclosure: I coauthored one of those reports.
34United States of America v. Internet Research Agency, LLC, et al., www.justice.gov/
opa/press-release/file/1035562/download.
35Ibid., p. 12.
36Ibid., p. 13.
37Ibid., p. 7.
84
Chapter 5 | Democracy Hacked, Part 1
ORGANIZATION employees, referred to as “specialists,” were
tasked to create social media accounts that appeared to be oper-
ated by U.S. persons. The specialists were divided into day-shift and
night-shift hours and instructed to make posts in accordance with
the appropriate U.S. time zone. The ORGANIZATION also circu-
lated lists of U.S. holidays so that specialists could develop and post
appropriate account activity. Specialists were instructed to write
about topics germane to the United States such as U.S. foreign pol-
icy and U.S. economic issues. Specialists were directed to create
“political intensity through supporting radical groups, users dissatis-
fied with [the] social and economic situation and oppositional social
movements. ”38
This included focusing messaging around themes like immigration, Black Lives
Matter and police brutality, Blue Lives Matter, religion, and regional secession,
among others. These topics were guided through internal documents provided
to IRA “specialists” to use as the basis of their content, and in September
2016, one internal memo stressed that “it is imperative to intensify criticizing
Hillary Clinton” in advance of the November election. 39
The internal documents made public by the Department of Justice in their
indictments of key IRA officials and shell companies only provide a small
window into their actual operations targeting Americans, though. To really
see what they did and how their content spread, the Senate Select Committee
on Intelligence commissioned two groups to analyze private data provided to
the Senate by Twitter, Facebook, and Google and report their findings. Several
of my colleagues and I had the honor of contributing to one of those reports.
While those datasets were all missing key metadata (and, I believe, further
examples of IRA and other Russian agencies’ U.S.-directed propaganda), they
exhibit a massive and professional operation that far exceeds the initial
statements made by platform executives. It is impossible to quantify how
many votes this campaign may have changed, or at least influenced, but it is
impossible to deny that this operation, in conjunction with Fancy Bear’s work,
was a significant factor in the tone of the election, the issues that took center
stage in public discourse, and the media coverage around the election. All of
this together certainly influenced some votes and has since cast doubt on the
legitimacy of the 2016 election and fear about foreign influence of previous
and subsequent elections.
But just what did their operation look like? And how pervasive was it?
38Ibid., p. 14.
39Ibid., p. 17.
Data versus Democracy
85
To start with, IRA influence operations around the 2016 U.S. election hit
every major platform, and even some minor ones. In addition to Facebook,
Instagram, Twitter, and YouTube, evidence of IRA operations has surfaced on
Google Plus, Vine, Gab, Meetup, VKontakte, LiveJournal, Reddit, Tumblr,
Pinterest, Medium, and even Pokémon Go.40 That’s to say nothing of the
world-wide web, where the IRA (and other branches or contractors of the
Russian government) have web sites, blogs, and pro-Kremlin “think-tank”
journals. This network of IRA web assets was “run like a sophisticated
marketing agency” with dozens of real people posting, sharing, retweeting,
and commenting on each other’s memes, blogs, and tweets. As my colleagues
and I wrote in our report for SSCI, “it was far more than only $100,000 of
Facebook ads, as originally asserted by Facebook executives. The ad
engagements were a minor factor in a much broader, organically-driven
influence operation. ”41 The overall budget for Project Lakhta exceeded $25
million, 42 which primarily went to paying employees to create not ads but
organic content: tweets, posts, memes, videos, events, all shared from user
accounts and pages belonging to fake personas and groups carefully crafted by
IRA “specialists.” Overall, IRA content pre- and post-election reached an
estimated 126 million Americans on Facebook, 20 million on Instagram, and
1.4 million on Twitter. This was no tiny operation.
It’s also important to note that the IRA, by and large, did not operate a
network of automated accounts, known as a botnet. IRA employees were
expected to meet daily quotas of organic posts, comments, shares, and likes.
These were mainly human-operated accounts that sought to “weave
propaganda seamlessly into what appeared to be the nonpolitical musings of an
everyday person.” 43 Thus, they had employee shifts that lined up with U.S. time
zones (see the DoJ indictment previously discussed) and a system of bonuses
and fines that encouraged individual “specialists” to produce high-engagement
content.44 To appear even more like real Americans, IRA “specialists” played
hashtag games45 and posted a high volume of nonpolitical content.
Project Lakhta involved far more than pro-Trump and anti-Clinton messages
blasted into the ether in the hopes of reaching an audience. In fact, election-
related posts only accounted for 7% of IRA Facebook content, 18% of
40Renee DiResta, Kris Shaffer, Becky Ruppel, David Sullivan, Robert Matney, Ryan Fox,
Jonathan Albright, and Ben Johnson, “The Tactics and Tropes of the Internet Research
Agency,” New Knowledge, published December 17, 2018, https://disinformationre-
port.blob.core.windows.net/disinformation-report/NewKnowledge-
Disinformation-Report-Whitepaper-121718.pdf, p. 5.
41Ibid., p. 33.
42Ibid., p. 6.
43Adrien Chen, “The Agency,” The New York Times Magazine, published June 2, 2015, www.
nytimes.com/2015/06/07/magazine/the-agency.html.
44Ibid.
45“The Tactics and Tropes of the Internet Research Agency,” p. 13.
86
Chapter 5 | Democracy Hacked, Part 1
Instagram content, and 6% of Twitter content. 46 Rather, the IRA created
what my colleagues and I call a “media mirage”—a false, interconnected,
multiplatform media landscape, targeting multiple different communities
with deceptive, manipulative messaging. 47 This “mirage” included a significant
portion of apolitical content, and where the content was political, it was
often focused on current divisive social issues that energized (or
de-energized) members of different communities, rather than specific
candidates. This mirage targeted three general communities—right-leaning
Americans, left-leaning Americans, and African Americans—as well as more
hyper-targeted subcommunities like pro-secessionist Texans, democratic
socialists, evangelical Christian
s, etc. And this mirage targeted them with
real news, fake news, disingenuous conversation, and—likely most
significant—meme warfare.
The IRA had done their homework—both online and on the ground in the
United States—when it came to targeting American communities. (And they
constantly retooled their messaging based on user engagement stats, just like
one would expect from a digital marketing firm.) In many cases, they targeted
communities with specific messages tailored for that community, which, of
course, fit the Kremlin’s agenda. For example, they targeted right-wing
Americans with narratives that would get them energized to come out and
vote against democratic or moderate candidates—fearmongering narratives
about immigration and gun rights, inspirational Christian-themed narratives,
and warnings about Clinton’s alleged corruption. They targeted left-leaning
Americans with narratives that would de-energize them, turning their Clinton
support lukewarm or encouraging a third-party or write-in protest vote—
narratives about Clinton’s alleged corruption, the DNC’s undemocratic
primary process that denied Bernie Sanders a fair shot, feminist and
intersectional narratives that labeled Clinton a bad feminist. And they targeted
African Americans with even more poignant voter-suppression narratives
about police brutality or the racist tendencies of both parties that were
intended to turn them off from the democratic process in general. The
ultimate goal was to encourage votes for Trump and—if not possible to flip a
leftist to vote Trump—to depress turnout from Democrats and from
demographics that tended to vote Democrat. And they pursued this goal
more through general social and political narratives than through posts that
referenced the candidates specifically.
Notice, though, that most of these narratives are about distrust and fear—
not positive American virtues that happened to be consistent with Russian
aims. Yes, there were some high-engagement patriotic accounts and pro-
Christianity accounts, but overall, even these accounts were about creating
an insider/outsider framework that could be operationalized in other
46Ibid., p. 76.
47Ibid., p. 45.
Data versus Democracy
87
contexts with narratives of fear or anger about outsiders. Many of the pro-