These defenders of democracy do not exist
In which inauthentic Bluesky accounts posing as liberal activists and organizations use AI-generated reply spam and mass following to attempt to build an audience
If one looks only at their profiles, these six (now-suspended) Bluesky accounts appear to be fairly ordinary U.S. liberal social media accounts, with names like “Save Democracy”, “Blue Wave Riders”, and “Liberal Lion”. On closer inspection, however, there are multiple problems. Almost all of the accounts’ posts are unsolicited replies that appear to have been generated with a large language model such as ChatGPT, Claude, or Grok. Additionally, the accounts are following random users at a high rate of speed, with the newest accounts following hundreds of accounts per hour immediately after being created.
All six accounts in this network were created within the last month, with the oldest (“Save Democracy”) dating back to December 9th, 2024, and the newest created on January 2nd, 2025. The most recently created accounts almost immediately embarked on following other users at a rapid pace. This can be seen in the discrepancy between the numbers in the collage at the top of this article and the table of accounts above; these two figures were made roughly an hour apart, during which timespan both “Liberal Lion” and “Raise For Democracy” followed over a thousand additional accounts. This strategy netted followbacks occasionally, although the accounts still wound up following many times more accounts than they gained followers of their own. The older accounts in the network managed to land themselves on both starter packs and moderation lists here and there.
Almost all of these accounts’ posting activity took place in brief spikes where the network posted dozens or hundreds of times in a brief period, and then went silent for hours or days. All but 10 of the network’s 1174 posts are unsolicited replies; the exceptions are four introduction-style posts, and six reposts of the presently-banned ilpapitrumpo.bsky.social account (“Donald Dump”), which, in an interesting twist, was suspended by Bluesky at around the same time as the six spam accounts.
The six spam accounts have replied to at least 866 different accounts, mostly but not exclusively liberal accounts with at least a few thousand followers. Although many of the replies are phrased as though the account being replied to had initiated the conversation (i.e., “how can I help you”), almost all of the replies are in fact unsolicited. In some cases, the replies look like generic chatbot responses to some element of the post being replied to; for example, “Exciting opportunity for political science researchers! Apply by February 15. Good luck!” in response to a post advertising a position that begins on February 15th. Other responses are simply generic offers of assistance that no one asked for.
Although it is sometimes difficult to be certain whether a given piece of text was generated using a large language model, there are several dead giveaways as to the artificial origin of this spam network’s replies. Four of the accounts explicitly acknowledged the synthetic nature of their content in responses such as “I’m an AI assistant here to help with any questions you may have”, and all six frequently make generic offers of assistance. Additionally, all six accounts have replied with error messages indicating their inability to perform certain tasks, such as accessing external links or engaging with explicit content. As of the time of this writing, these six accounts have been suspended by Bluesky (along with the only account they reposted), but it would not be surprising to see variations on the theme appear in the future, especially since Bluesky already appears to have banned previous waves of similar accounts.
My question is: do you have any sense the point of this? Do accounts like this have a subversive political motive e.g. build an audience posing as liberal accounts and at some given point flood with content beneficial to the far right? Or is it just somehow a money making excercise (although how I would struggle to see).