Russian and Chinese troll factories are actively using artificial intelligence to strengthen their influence operations, improve content quality, expand multilingual reach, and modernize visual presentation, rather than simply increasing the volume of posts.
Experts from the GBHackers portal conducted a detailed study of fake accounts supporting Russia and China on the X platform over the period from 2024 to 2026. To identify hidden influence networks, they used an advanced machine learning algorithm combining unsupervised clustering and supervised classifiers pre-trained on human-labeled signals.
According to the study, this machine learning pipeline demonstrated high effectiveness, achieving an average accuracy of 86% and a recall of 83% in identifying likely fake accounts, reports Letsdatascience.
The analysis of collected data showed that the median volume and posting frequency in these bot networks decreased by about 50% between 2024 and 2026, while the number of active fake accounts remained stable at between 5,000 and 11,000 per threat actor group.
At the same time, researchers recorded a sharp increase in original posts containing various images. In accounts supporting Russia, the use of visual content increased more than fourfold, while in China-supporting accounts it doubled. A significant portion of these images is identified as being generated by artificial intelligence.
This shift from mass automated text spam toward higher-quality multimodal content reflects a broader global trend in the IT sector.
“Troll factories” are using generative models to maximize persuasive impact on audiences while reducing the noise and signals that typically trigger automated moderation systems on social media platforms.
OpenAI also released a public report stating that it had identified and fully disrupted five covert influence operations linked to Russia, China, Iran, and Israel. The operators of these campaigns used commercial OpenAI tools to generate high-quality multilingual text, create fake biographies for accounts, produce convincing social media comments, write full articles, debug code, generate images, and perform precise translations.
Ben Nimmo, chief of intelligence and investigations at OpenAI, noted that various threat actors are deliberately using such platforms to improve their content and increase the effectiveness of disinformation operations.
