We've seen some glitchy GAN-generated images used as Twitter profile pics before, but @jtatejtate1 kinda takes the cake with the utterly surreal "clothing" and "hat". #SaturdayShenanigans
Here's a video showing @jtatejtate1's profile pic blended with a bunch of other GAN-generated images. The major facial features (eyes, nose, and mouth) are in the same location on each image.
(more on GAN-generated face pics and the usage thereof on Twitter accounts)
The GAN-generated image isn't the only thing that isn't human about @jtatejtate1. Its tweets appear to be automated based on two pieces of evidence. First, the app used to send its tweets has a blank name (as opposed to Twitter Web App/Twitter for iPhone/Twitter for Android/etc).
Second, there's a timing pattern in its original tweets (non-retweets): nearly all of them were sent one second into the minute they were tweeted. This is likely the result of the tweets being sent on a timer, with an average of one second of lag to reach Twitter's servers.
All of @jtatejtate1's tweets subsequent to August 14th, 2020 are retweets. The majority of them contain "music" or related terms. Its two most recent tweets are exceptions; it's possible it switched to keying off the word "algorithm" but too soon to tell.
Who follows @jtatejtate1? Weirdly, 31 of its followers are batch-created accounts from September 2013 that went on a 7 year hiatus and have now reactivated and started tweeting again. (@jtatejtate1 itself was created in July 2020).
By exploring the follower networks of those 31 accounts, we found a group of 119 accounts that we believe to be part of a bulk follow network, all created on September 26th 2013. These accounts were dormant from late 2013 until mid-2020, but are now tweeting and following again.
Who does this network follow? By and large, Arabic-language accounts. A notable exception is @freitagusa, a followback #MAGA account. Many of the accounts followed by the network have 2020 creation dates, indicating the follows are recent rather than from 2013.
The content tweeted by this network since it reawakened consists mostly of quotes plagiarized from a variety of sources. This doesn't give much of a clue as to its purpose, but we'll keep an eye on it and see what it does going forward.
Meet @SRRRJ, @Srm1n, @DemiLovatoTH, @banci__, and @gdibarry, a quintet of automated accounts that tweet CNN articles accompanied by partial headlines and tag @null (a suspended account) in each tweet.
These five accounts presently send all of their tweets via automation service twittbot(dot)net. There are two distinct schedule patterns, so it's possible this is two separate botnets, but since the accounts are otherwise similar we analyzed them as a single network.
What does this botnet tweet? It's quite single-minded: all recent tweets (the last ~3200 from each account) contain links to CNN, accompanied by the beginning of the title of whatever article/video is being linked.
We've analyzed a variety of bot/sock networks using coordinated retweeting to make content look more popular than it actually is (astroturfing). For a change of pace, here's a look at a network from 2018 that used quote tweets for artificial amplification.
This June 2018 tweet from @OrigoNetworks received more than 3 times as many quote tweets as retweets (quote tweet/retweet ratio of ~0.05 appears to be average). Furthermore, the accounts that quote tweeted it were disproportionately created in June/July 2018. What happened?
To explore further, we looked at other tweets with lopsided quote tweet/retweet ratios quote tweeted by the bulk-created accounts that quote tweeted the @OrigoNetworks tweet, and indeed found more accounts created in June/July 2018 that quote tweeted a lot of the same tweets.
If what's missing from your Wednesday is a group of newly-minted automated accounts that link randomly named NFL and NBA-themed tumblr pages, this is the botnet for you. (We're as confused as you as to what 1990s NASCAR drivers have to do with the Lakers.)
We found 29 accounts, all created between November 8th and November 11th 2020. These accounts post all of their tweets via automation service dlvr(dot)it.
Most of this botnet's tweets contain links to tumblr pages of the form nfl*****(dot)tumblr(dot)com or lakers*****(dot)tumblr(dot)com. The tumblr pages are empty other than a "RELOAD NEWS" button that redirects to zeezhop(dot)com. (As always, be wary of links to unknown sites.)
How did this duo of newly-created IFTTT bots get so many retweets on their tweets with embedded ISIS videos despite having almost no followers? #TuesdayThoughts
(Some of the accounts mentioned in this thread got suspended while we were doing the research.)
We downloaded the retweets of the video tweets from the two IFTTT ISIS bots (@3J2IEgC35AMih8R and @Id1kveiZWhI013x), and it turns out that most of the retweets are also automated: 250 of 257 retweets (97.3%) were sent via one of 101 different automation apps.
The accounts retweeting the two IFTTT ISIS bots didn't only automate their retweets of these two accounts: 89 of 122 accounts (72.9%) automated at least half of their tweets, and 65 of 122 (53.3%) are at least 80% automated. These accounts were mostly created in 2020.
Meet @Wizard_Predicts, a self-anointed Election Wizard with 106K followers and a taste for dubious tweets about voter fraud. It takes pride in being followed by @DanScavino, @MattGaetz, @DLoesch, and many more. (We'll get back to the "many more" part shortly).
The first unusual thing we noticed about @Wizard_Predicts: despite it being an English-language pro-Trump account, many of the "Who To Follow" suggestions that come up when viewing its timeline are non-English accounts.
Although @Wizard_Predicts's followers mostly look like English-language #MAGA accounts, things get a weirder when we look at the first 10K or so: the 2009 spike in account creations we usually see is largely absent. (Its first follower is right-wing Twitter pundit @Barnes_Law.)
We found 25574 automated accounts that we believe to be part of this botnet. All were created in batches between 2012 and 2014, and they use repetitive naming schemes. (This is probably not the whole network, we stopped searching when the pace of finding more accounts slowed.)
These accounts are mostly dormant, but back when they were active they tweeted via a wide variety of automation apps (883 apps total). Almost all of their content is in Russian.