Thread by @0rf

1/ #TWITTERFILES 21
How to Find Russians Anywhere

Pt.1 - PROJECT OSPREY -

After the 2016 election, the Senate Intel Committee asked Twitter to identify accounts from Russia's Internet Research Agency. But both Twitter and 3rd party researchers struggled to find Russian agents.

2/ At first, Twitter was only able to produce 22 plus 179 IRA "linked" accounts. Democrat Mark Warner called the numbers "inadequate on every level". Twitter went back to the drawing board in an analysis project called "Osprey".

3/ In "Project Osprey," Twitter counted 2 types of Russians.

1. "A Priori" Russians - identified as Russian by outside researchers like QIntel

2. "Inferred" Russians (aka is_russian) - identified by algorithm tracking "signals" like Cyrillic text or a Russian IP address

4/ This kind of analysis, based on "markers" like type of email carrier or retweeting history, can quickly become a Rorschach test, where you see what you want to see. "If you just look for that marker, then everything will look Russian," is how one industry analyst puts it.

5/ Even Twitter understood. Noting Green Party candidate @DrJillStein was "captured by is_russian" - the "inferred Russians" list - Twitter analysts commented on the "overly broad nature of is_russian."

6/ "Outrageous," says Stein. "Just more that Joe McCarthy would be proud of."

7/ Another account deemed "is_Russian" was @Wikileaks. Shown the attribution, @Stella_Assange said, "Wikileaks does not fit the definition," noting that anyone using surveillance-resistant search tools like the Tor Browser might also be deemed "Russian."

8/ "Use of TOR would randomly result in false positives," Stella Assange said. "TOR is an essential tool for civil liberties and privacy communities."

9/ In the same Osprey document, Twitter notes the hashtag #WarAgainstDemocrats (which NYT reported was posted a scary 1,700 times on election day) only garnered a microscopic 6,953 impressions—a hydrogen atom in Twitter's vast ocean of tweets.

10/ After Twitter's early attempts to identify Russian accounts resulted in such low numbers, they used different methodologies, tallying ever-increasing numbers of "Russians," from 22 to 201 to 2700 to 2,752 to 3,124 (in Osprey), & finally, 3,814.

11/ Even as Twitter began referring openly to "IRA-linked" accounts, the company had no sure way to make such identifications. They were only "potentially connected".

12/ As policy chief Colin Crowell explained in late 2017, "we are citing accounts as IRA on the basis of 3rd party assessments," as "we have no realistic way of knowing this on a Twitter-centric basis," apart from "educated guesses."

13/ Twitter knew that if they couldn't identify IRA accounts with certainty, outside researchers lacking Twitter's internal data couldn't either.

PART 2 - Russian Troll Hunters

"They just don't have the chops"

14/ If you look on today, you'll see that MSNBC alone made hundreds of false claims of Russian meddling, citing the Hamilton 68 dashboard, which past #TwitterFiles revealed to be tracking Americans, not "Russian bots and trolls," as claimed. Racket.News

15/ Outside of Hamilton 68, about which Twitter stayed quiet despite internal knowledge that it was "bullshit," Twitter analysts may have had the most disdain for Clemson's Media Forensics Hub, a major driver of print and TV cyber-scare stories of Russian subversion.

16/ Clemson's "Troll Hunter" profs Darren Linvill & Patrick Warren appeared in one media story after another (sometimes photographed w Hamilton 68 dashboards in the background), even warning that "uplifting tweets" could lead you into the dark clutches of a Russian troll.

17/ When asked how they knew an account was Russian. Warren confessed he can't be "100% sure" but Livill said he was "certain". He wouldn't say how though. "Transparency just isn't possible," he said.

18/ Clemson's most "certain" troll hunter seemed to jump the gun in 2020 when he suggested #BloombergIsRacist could be a “Russian hashtag” bc it started that morning from the account @drkwarlord that “live[s] somewhere in Asia” bc it posts when Americans should be sleeping.

19/ That was false: It started prior to that morning, and not by @drkwarlord, who is an American living in the US, not Asia. When I asked why his sleep sched seemed off, he laughed, and told me, “I’m a nurse at a hospital in Indiana. In 2020, I worked the night shift.”

20/ This kind of mixup appears to be what Twitter's Aaron Rodericks was referring to when he noted the Clemson professors' "constant tendency to find Russians/foreign interference every time they look for anything."

21/ Another example occurred w the #DCBlackout hoax. As Twitter exchanged emails with the FBI, describing the campaign as "a small-scale domestic troll effort...not a significant bot or foreign angle,” Linvill described the campaign as having employed “a classic Russian move".

22/ After a later Clemson report regarding "inauthentic activity" during Vice Presidential debates, Rodericks decried "their usual unsubstantiated claims of foreign interference," writing they had to "scrape the bottom of the barrel to find foreign interference".

23/ But to the Clemson researchers' credit, they found Twitter was "scraping the barrel" too, having looped in several Americans in a list of "Russian bots" given to Congress. So clearly, Twitter's own false positives didn't stop with its project, Osprey.

24/ The Americans found by Linvill & Warren had been unjustly censored—suspended from Twitter without warning or explanation. Thrown in a haystack of suspected Russians, "they lost access to Twitter accounts that they used to maintain social and career connections," wrote Wired.

25/ After years of Red Scare stories (citing the Clemson duo, Ham68, & others) that fueled such awful censorship, Linvill leveled with PBS in 2022: "You know, Russian trolls aren't as common as people think they are."

26/ PART 3 - HOW TO FIND DISINFORMATION EVERYWHERE

To be continued...

#TwitterFiles21 co-authored w @mtaibbi and help from other #TwitterFiles researchers

@mtaibbi 27/ The Clemson professors' comments to Racket News

Note: I cited Osprey for one tally (3,124) which comes from the Osprey version I've covered here in #TwitterFiles21 *But theoretically, all these numbers could have come from project "Osprey" analyses—just different versions that I haven't seen...yet. The documents I had access to are limited. Happy to look through more if I were to have the opportunity @ElonMusk.

Corrections:

The last number in the sequence is 3841 (not 3814)

Seems the order went from
22 to 201 to 3,124 to 2700 to 2,752 to 3,841

So not every tally went up but it usually did

"But they are there," he continued...

21/ #TwitteFile email referenced: