This war correspondent does not exist
A rogue's gallery of renamed Twitter accounts with artificially generated faces masquerading as journalists covering the war in Ukraine
Russia’s 2022 invasion of Ukraine has been accompanied by widespread interest in real-time news and information on the conflict, and an explosion of Ukraine-themed “news” accounts on social media have emerged, purportedly to provide coverage. Healthy skepticism of accounts of this sort is warranted, however, as many are not as they seem. One popular tactic: creating a “journalist” by renaming an old Twitter account, slapping an artificially generated or stolen/stock photo on it, and tweeting plagiarized content or outright falsehoods as “breaking news”. Here are three examples of “journalist” accounts made in this manner, all of which use GAN-generated faces. (GAN = “generative adversarial network”, the machine learning technology behind tools like https://thispersondoesnotexist.com and https://generated.photos.)
In March 2022, a supposedly Ukraine-based Twitter “journalist” by the name of @BethBoykins22 with a GAN-generated face started tweeting out news about the war in Ukraine, usually with accompanying videos. None of @BethBoykins22’s “reporting” was original, however — both the videos and the tweet text were copied verbatim from a mix of public figures such as Alexander Vindman and random “breaking news” Twitter accounts (some of which likely also plagiarized the content).
Every account on Twitter has a permanent numeric account ID that remains constant even if the account’s handle, display name, or biography is changed. This can be used to confirm suspected handle changes. In the case of @BethBoykins22, this numeric ID is 860123802684051456, and a search of the Wayback Machine for archives of one of the account’s tweets reveals that as recently as February 21st, 2022, the account was named @iflehbrahim7 and claimed to reside in Arkansas rather than Ukraine.
Tweets also have numeric IDs (they’re even part of the tweet URL), and much like the account ID, they remain the same when the account is renamed.
On March 2nd, 2022, a Twitter account named @lubadovzhenko1 with a GAN-generated face began tweeting about the war. Although the account operator claimed to be a journalism student with poor English skills and no previous Twitter experience, this is far from accurate; the account was simply renamed and purged of old tweets. Wayback Machine confirms that the @lubadovzhenko1 account (numeric ID 1357665122395815940) was named @camplostkids prior to the invasion of Ukraine, and several of the deleted tweets show fluent usage of conversational English.
This particular fake journalist account even managed to get a guest column in a major UK media outlet. On March 17th, 2022, The Times published a piece on wartime life in Kyiv allegedly written by @lubadovzhenko1 (as “Luba Dovzhenko”) and featuring the same GAN-generated face as the Twitter account. After the @lubadovzhenko1 account was exposed as inauthentic and covered by France 24, The Times belatedly removed the guest article.
In mid-2022, an “Open Source Intelligence” Twitter account with the handle @na_intel and a GAN-generated face from https://generated.photos went viral by “reporting” on the war in Ukraine via plagiarized photos from official Ukrainian military Facebook pages. As with the two other accounts discussed in this article, the @na_intel account (numeric ID 873220123) has been renamed at least once. In an interesting twist, this account has used multiple GAN-generated faces: immediately prior to being renamed to @na_intel, it was named @kristoferso and bore a different GAN-generated face, and has also repeatedly promoted a Patreon account featuring yet another GAN-generated face (also obtained from https://generated.photos).
How can we tell that the “faces” these three accounts used as profile images are artificially generated? There are multiple signs, including but not limited to:
deformed microphone that melts into @BethBoykins22’s clothing and face
surreal clothing on @BethBoykins22’s image
boundaries between hair and background are unrealistic in areas containing loose strands of hair
@na_intel’s ears do not match
Additionally, GAN-generated faces in wide circulation (including those used by these three accounts) have the telltale trait that prominent facial features are drawn in the same location on every image, regardless of what direction the “face” appears to be looking in. For example, the eyes are always horizontally positioned halfway between the center and edge of the image, and vertically 7/8 the distance from top to center. (@na_intel’s profile image has been cropped and does not have this property, but since the image provably originated with https://generated.photos and has other weird artifacts, we can safely conclude that it is GAN-generated.)
These three accounts have all been suspended by Twitter, but they are by no means the only examples of fake “journalist” accounts on Twitter (or other social media platforms) posting about the war in Ukraine or other topics. The methods they utilized — GAN-generated faces, recycling old accounts rather than creating new ones, and building an audience by presenting plagiarized media content as breaking news — continue to be commonplace. This type of deception is particularly concerning on Twitter at present in light of ongoing plans to include a blue verification checkmark with the Twitter Blue monthly subscription service, as inauthentic “journalist” accounts accompanied by blue checks are more likely to be seen as legitimate than the accounts describe in this article.
The research in this article was done by myself and ZQ (@ZellaQuixote on Twitter) and originally presented in these three Twitter threads: