The article reports how the Russian propaganda network “Pravda” implemented “LMM grooming” by producing 3.6M articles promoting russian propaganda narratives. The authors asked 10 different large LMM models 15 questions regarding the content of those articles. In 33% of cases, LMM cited Pravda network or included it in the list of trustworthy sources. The researchers found that Pravda specifically targeted LMMs since its unique user count stays under 100. The Pravda network was founded after the Russian full-scale invasion and increased the number of its domains to 200 in 2024. The article shows how LLMs are vulnerable to data poisoning and how propaganda can use it to spread false narratives.
