26.05.2025
2 min
863
Former US Marine John Mark Dougan used Claude AI to create tens of thousands of pro-Russian articles targeting American audiences. Leaked files from his laptop revealed how the propagandist attempted to influence not only readers but also the responses generated by other AI chatbots.
Dougan used Anthropic’s Claude to prepare content designed to appear credible to American readers. The AI-generated articles were published on specially created websites to spread narratives favorable to Russia and make them accessible to search engines and artificial intelligence models.
Details of the operation emerged following a leak of Dougan’s emails, Telegram and Signal messages, and AI prompts. The stolen materials were published on a website impersonating his own platform, while links to the files were distributed through Dougan’s compromised X account. Dougan himself confirmed to NewsGuard journalists that the leaked materials were authentic. However, he said he was not concerned about the leak or what others might think of him.
One of the leaked documents revealed a strategy aimed at influencing search algorithms and large language models (LLMs). Dougan regularly updated pro-Russian news articles using AI, attempting to improve their readability and search engine rankings. The goal was to get these materials into search results, where other chatbots might use them when generating responses.
The scale of the operation was substantial. According to the leak, Dougan’s team produced approximately 43,000 articles per month, at a cost of around 2 million rubles, or $23,000 monthly. In one of his prompts to Claude, he requested a breakdown of expenses, noting that producing such a large volume of content was becoming too expensive. The documents also revealed that the propagandist used not only Claude but also Alibaba’s Chinese chatbot Qwen and Open WebUI, a platform that provides a unified interface for working with different AI models.
In one instance, Dougan asked Claude to help create a website about freedom of speech for a European audience, specifically targeting French-speaking users. He considered the name SpeakUp but questioned whether French audiences would understand it. The chatbot explained why the name might not be suitable for the intended audience. Using Claude to spread misleading content violates Anthropic’s policies, which prohibit creating disinformation, fake identities and accounts, and falsely attributing authorship. Despite similar restrictions imposed by Alibaba, Dougan also asked Qwen how to create content that would evade AI detection systems.
The mere use of these chatbots does not prove that they complied with all of Dougan’s requests or that his content actually influenced the responses of other AI models. However, the leak revealed the extent of his efforts to automate the production and distribution of propaganda. John Mark Dougan, also known by the alias BadVolf, previously served in the US Marine Corps and worked as a deputy sheriff in Palm Beach County, Florida, and later in Windham, Maine. He was dismissed over disciplinary violations and harassment.
In 2016, Dougan became the subject of an FBI investigation involving the leak of confidential police documents. After facing 21 felony charges in Florida, he fled to Russia, where he received political asylum and became a Russian citizen in 2023. Previous reports have also alleged links between Dougan and Russia’s military intelligence agency, the GRU, which reportedly financed his propaganda activities. Dougan created and operated fake news websites that impersonated legitimate American media outlets and spread disinformation among US residents.
The leak demonstrated how generative AI can be used to mass-produce politically motivated content and attempt to influence the information environment. However, the actual impact of this campaign on other chatbots’ responses remains unknown.
