Twitter Updates Info on Propaganda, Internet Research Agency Linked Accounts, Russian Bots

Most user engagement was with a very small number of IRA-associated accounts.  Image above courtesy Twitter, example of Internet Research Agency disinformation campaign.
By Twitter PublicPolicy

Friday, 19 January 2018

When we appeared before the United States Congress last fall, Twitter publicly committed to regularly updating both congressional committees and the public on findings from our ongoing review into events surrounding the 2016 U.S. election.

Twitter is committed to providing a platform that fosters healthy civic discourse and democratic debate.  We have been cooperating with congressional investigations into Russian interference in the 2016 U.S. presidential election. We have committed to be as transparent as possible about sharing what we have learned through our retroactive investigationinto activity related to the election.

Since we presented our findings to Congress last fall, we have updated our analysis and continue to look for patterns and signals in data. Today, we are sharing an update on several aspects of that ongoing work, as well as steps we are taking to continue to make progress against potential manipulation of our platform.

Informing People of Malicious Activity in the 2016 Election

As previously announced, we identified and suspended a number of accounts that were potentially connected to a propaganda effort by a Russian government-linked organization known as the Internet Research Agency (IRA).

Consistent with our commitment to transparency, we are emailing notifications to 677,775 people in the United States who followed one of these accounts or retweeted or liked a Tweet from these accounts during the election period. Because we have already suspended these accounts, the relevant content on Twitter is no longer publicly available.

Examples of IRA Content

Most user engagement was with a very small number of IRA-associated accounts.  Some examples of content which received significant engagement are:

Updated Numbers of IRA Accounts

As part of our ongoing review, we have identified both more IRA and automated Russia-based accounts. The results of this supplemental analysis are consistent with the results of our previous work: automated election-related content associated with Russian signals represented a very small fraction of the overall activity on Twitter in the ten-week period preceding the 2016 election.

We have identified an additional 1,062 accounts associated with the IRA. We have suspended all of these accounts for Terms of Service violations, primarily spam, and all but a few accounts, which were restored to legitimate users, remain suspended.  At the request of congressional investigators, we are also sharing those account handles with Congress. In total, during the time period we investigated, the 3,814 identified IRA-linked accounts posted 175,993 Tweets, approximately 8.4% of which were election-related.

We have also provided Congress with the results of our supplemental analysis into activity believed to be automated, election-related activity originating out of Russia during the election period. Through our supplemental analysis, we have identified 13,512 additional accounts, for a total of 50,258 automated accounts that we identified as Russian-linked and Tweeting election-related content during the election period, representing approximately two one-hundredths of a percent (0.016%) of the total accounts on Twitter at the time.  However any such activity represents a challenge to democratic societies everywhere, and we’re committed to continuing to work on this important issue.

Enhancing Information Quality

After the 2016 election, we launched our Information Quality initiative to further develop strategies to detect and prevent bad actors from abusing our platform. We have since made significant improvements, while recognizing that we have more to do as these patterns of activity develop and shift over time.

With our current capabilities, we detect and block approximately 523,000 suspicious logins daily for being generated through automation. In December 2017, our systems identified and challenged more than 6.4 million suspicious accounts globally per week— a 60% increase in our detection rate from October 2017. We have developed new techniques for identifying malicious automation (such as near-instantaneous replies to Tweets, non-random Tweet timing, and coordinated engagement). We have improved our phone verification process and introduced new challenges, including reCAPTCHAs to validate that a human is in control of an account.

Alongside these improvements, we’re continuing to expand enforcement of our developer and automation rules. Since June 2017, we’ve removed more than 220,000 applications in violation of our rules, collectively responsible for more than 2.2 billion low-quality Tweets.

In 2018, we will build upon our existing improvements. Our plans include:

  • Investing further in machine-learning capabilities that help us detect and mitigate the effect on users of fake, coordinated, and automated account activity;
  • Limiting the ability of users to perform coordinated actions across multiple accounts in Tweetdeck and via the Twitter API;
  • Continuing the expansion of our new developer onboarding process to better manage the use cases for developers building on Twitter’s API. This will help us improve how we enforce our policies on restricted uses of our developer products, including rules on the appropriate use of bots and automation.

Media Literacy and Partnerships

We recognize that Twitter is an important part of a larger ecosystem of how news and information spreads online, and that we have a responsibility to support external programs that empower our users, connecting them with resources to give them control over their online experience.

Our partners Common Sense Media, the National Association for Media Literacy, the Family Online Safety Institute and Connect Safely, amongst others, have helped us to craft materials and conduct workshops to help our users learn how to process online information and understand which sources of news have integrity. We focus on elements like verification of sources, critical thinking, active citizenship online and the breaking down of digital divides.

Learn more about our most recent efforts for Media Literacy Week in countries like the U.S., Canada and Ireland, and follow our partners @MediaLiteracyEd@CommonSenseEdu and @ConnectSafely for new initiatives like the Teachers Institute at Twitter HQ.

Twitter is proud to partner with journalistic NGOs for trainings and outreach initiatives, including Reporters without Borders, the Committee to Protect Journalists, and the Reporters Committee for Freedom of the Press. We will keep working with reporters, journalism NGOs, and media organizations to ensure that Twitter’s full capabilities are built into newsrooms and established media outlets worldwide.

Moving Forward

Even as we continue to learn from the events of the 2016 U.S. election, we are taking steps every day to improve the security of our platform and stay one step ahead of those who would abuse it. As part of our preparations for the U.S. midterm elections, our teams are organizing to:

  • Verify major party candidates for all statewide and federal elective offices, and major national party accounts, as a hedge against impersonation;
  • Maintain open lines of communication to federal and state election officials to quickly escalate issues that arise;
  • Address escalations of account issues with respect to violations of Twitter Rules or applicable laws;
  • Continually improve and apply our anti-spam technology to address networks of malicious automation targeting election-related matters; and
  • Monitor trends and spikes in conversations relating to the 2018 elections for potential manipulation activity.

We are committed to ensuring that Twitter is safe and secure for all users and serves to advance healthy civic discussion and engagement. Our work on these issues will never be done, and we will continue in our efforts to protect Twitter against bad actors and networks of malicious automation and manipulation.

Milo Yiannopoulos, Roger Stone Announce Anti-Trust Lawsuit Against Twitter

Award-winning journalist and New York Times bestselling author MILO is joining longtime Donald Trump advisor and Republican strategist Roger Stone in an anti-trust lawsuit against social media giant Twitter.

In July 2016 MILO made headlines when Twitter permanently suspended his account following a negative review he wrote of the newly released, all-female Ghostbusters movie.

Stone had his account permanently suspended in late 2017 after posting several tweets with expletives. Both contend they did not violate Twitter’s terms of service and were targeted instead for arbitrary reasons.

In a statement about MILO joining the lawsuit, Roger Stone said:

We continue to explore a broad lawsuit challenging Twitter’s censorship and the hypocrisy of their rules for online conduct which seem to be disproportionately levied against conservative voices in an obvious attempt to silence us. Verified tweeters call for my murder online every day, but Twitter doesn’t ban them.

We believe it is time to expose their manipulation of algorithms, ‘shadowbanning’ and other online techniques used to limit our reach. It’s time for Twitter to be regulated like a public utility or perish. I am heartened that my friend Milo is prepared to join our legal action along with other conservatives who have been gagged by the Twitter censors.

In a January 11 statement about the anti-trust case, MILO said:

I am Patient Zero of the Twitter war against conservatives and libertarians. The company declared war on free speech when it banned me in July 2016. At the time, I appreciated the free press. But I have come to realize that Twitter’s recklessness and bias toward conservatives and free thinkers represents a threat to free speech and democracy, such is Twitter’s monopolistic grip on journalistic discourse.

Footage released this week by investigative journalists at Project Veritas reveal a defiantly biased company whose hateful and divisive political attitudes are robbing libertarians and conservative journalists and media personalities of the right to freely express their opinions in the press.

The biggest tech debate of the next decade is whether technology companies, in particular social networks, should be regulated as public utilities. It is becoming increasingly clear, given their rampant abuses, that they should. And Twitter is the worst offender of them all.

Twitter Shadow Banning Undercover Video Released by Project Veritas

MILO discussed details of the case in an episode of his new show, THE MILO SHOW, located at dangerous.com, which first aired live on January 11, 2018.

Twitter Shadow Banning Undercover Video Released by Project Veritas

Steven Pierre, Twitter engineer explains “shadow banning,” says “it’s going to ban a way of talking”

Former Twitter software engineer Abhinav Vadrevu on shadow banning: “they just think that no one is engaging with their content, when in reality, no one is seeing it”

Former Twitter Content Review Agent Mo Norai explains banning process: “if it was a pro-Trump thing and I’m anti-Trump… I banned his whole account… it’s at your discretion”

When asked if banning process was an unwritten rule, Norai adds “Very. A lot of unwritten rules… It was never written it was more said”

Olinda Hassan, Policy Manager for Twitter Trust and Safety explains, “we’re trying to ‘down rank’… shitty people to not show up,” “we’re working [that] on right now”

“Shadow banning” to be used to stealthily target political views- former Twitter engineer says, “that’s a thing”

Censorship of certain political viewpoints to be automated via “machine learning” according to Twitter software engineer

Parnay Singh, Twitter Direct Messaging Engineer, on machine learning algorithms, “you have like five thousand keywords to describe a redneck…” “the majority of it are for Republicans”

————–

(San Francisco) In the latest undercover Project Veritas video investigation, current and former Twitter employees are on camera explaining steps the social media giant is taking to censor political content that they don’t like.

This video release follows the first undercover Twitter exposé Project Veritas released on January 10th which showed Twitter Senior Network Security Engineer Clay Haynes saying that Twitter is “more than happy to help the Department of Justice with their little [President Donald Trump] investigation.” Twitter responded to the video with a statementshortly after that release, stating “the individual depicted in this video was speaking in a personal capacity and does not represent of speak for Twitter.” The video released by Project Veritas today features eight employees, and a Project Veritas spokesman said there are more videos featuring additional employees coming.

On January 3rd 2018 at a San Francisco restaurant, Abhinov Vadrevu, a former Twitter Software Engineer explains a strategy, called “shadow banning,” that to his knowledge, Twitter has employed:

“One strategy is to shadow ban so you have ultimate control. The idea of a shadow ban is that you ban someone but they don’t know they’ve been banned, because they keep posting and no one sees their content. So they just think that no one is engaging with their content, when in reality, no one is seeing it.”

Twitter is in the process of automating censorship and banning, says Twitter Software Engineer Steven Pierre on December 8th of 2017:

“Every single conversation is going to be rated by a machine and the machine is going to say whether or not it’s a positive thing or a negative thing. And whether it’s positive or negative doesn’t (inaudible), it’s more like if somebody’s being aggressive or not. Right? Somebody’s just cursing at somebody, whatever, whatever. They may have point, but it will just vanish… It’s not going to ban the mindset, it’s going to ban, like, a way of talking.”

Olinda Hassan, a Policy Manager for Twitter’s Trust and Safety team explains on December 15th, 2017 at a Twitter holiday party that the development of a system of “down ranking” “shitty people” is in the works:

“Yeah. That’s something we’re working on. It’s something we’re working on. We’re trying to get the shitty people to not show up. It’s a product thing we’re working on right now.”

Former Twitter Engineer Conrado Miranda confirms on December 1st, 2017 that tools are already in place to censor pro-Trump or conservative content on the platform. When asked whether or not these capabilities exist, Miranda says, “that’s a thing.”

In a conversation with former Twitter Content Review Agent Mo Norai on May 16th, 2017, we learned that in the past Twitter would manually ban or censor Pro-Trump or conservative content. When asked about the process of banning accounts, Norai said, “On stuff like that it was more discretion on your view point, I guess how you felt about a particular matter…”

When asked to clarify if that process was automated Norai confirmed that it was not:

“Yeah, if they said this is: ‘Pro-Trump’ I don’t want it because it offends me, this, that. And I say I banned this whole thing, and it goes over here and they are like, ‘Oh you know what? I don’t like it too. You know what? Mo’s right, let’s go, let’s carry on, what’s next?’”

Norai also revealed that more left-leaning content would go through their selection process with less political scrutiny, “It would come through checked and then I would be like ‘Oh you know what? This is okay. Let it go.’”

Norai explains that this selection process wasn’t exactly Twitter policy, but rather they were following unwritten rules from the top:

“A lot of unwritten rules, and being that we’re in San Francisco, we’re in California, very liberal, a very blue state. You had to be… I mean as a company you can’t really say it because it would make you look bad, but behind closed doors are lots of rules.”

“There was, I would say… Twitter was probably about 90% Anti-Trump, maybe 99% Anti-Trump.”

At a San Francisco bar on January 5th, Pranay Singh details how the shadow-banning algorithms targeting right-leaning are engineered:

“Yeah you look for Trump, or America, and you have like five thousand keywords to describe a redneck. Then you look and parse all the messages, all the pictures, and then you look for stuff that matches that stuff.”

UNDERCOVER VIDEO: Sr Network Security Engineer Reveals Twitter Ready to Give Trump’s Private DMs to DOJ

When asked if the majority of the algorithms are targeted against conservative or liberal users of Twitter, Singh said, “I would say majority of it are for Republicans.”

Project Veritas founder James O’Keefe believes the power over speech Silicon Valley tech giants has is unprecedented and dangerous:

“What kind of world do we live in where computer engineers are the gatekeepers of the ‘way people talk?’ This investigation brings forth information of profound public importance that educates people about how free they really are to express their views online.”

Project Veritas plans to release more undercover video from within Twitter in the coming days.

Mr. O’Keefe has just completed a book about this series entitled “AMERICAN PRAVDA: My fight for Truth in the Era of Fake News.” The book will be released by St. Martin’s Press on January 16, 2018.