Corrupt Comment: Anti-corruption foundation criticism

This case study was originally published in OpenAI’s October 2024⁠(opens in a new window) report.

Actor

We banned a small cluster of activity using our API to generate comments that were then posted on X. These comments typically criticized members of the Anti-Corruption Foundation, founded by the late Alexei Navalny. Given its focus on social media commenting and on making accusations of corruption, we call this operation “Corrupt Comment”.

Behavior

This actor used our models to generate English-language comments that were then posted by fake accounts on X—often in reply to posts that were in Russian. Although these comments were posted as replies, they were not generated as replies: we saw no indication that the operation was using our models to read or analyze social media content or come up with relevant responses. The activity was narrowly focused on senior figures at the Anti-Corruption Foundation (known as FBK for its Russian initials), founded by the late Russian activist Alexei Navalny. The operator or operators used our API to generate comments that criticized the Foundation, its leadership, and Navalny’s associates. Those comments were then posted on X in reply to posts by members of the Foundation. The X accounts which posted the comments were mostly created in December 2023, had zero followers, and did not receive any replies to their posts.

Some of the accounts on X featured profile pictures of scenery, copied from across the internet. Others used fake profile pictures that bore indicators of having been created using an earlier era of AI, generative adversarial networks (GAN)—a technique reported on as far back as 2019.

Completions

The comments that this operation generated primarily criticized the FBK, questioning its transparency. As with the Rwandan case, this approach is best considered as “theme and variations”, where many different posts were used to pass the same essential message.