Leaked documents reveal how former US Marine tried to use AI Claude to spread Russian ...

Leaked files from the laptop of former US Marine turned Russian propagandist John Mark Dugan have revealed the extent of his disinformation work. The documents show that he systematically used artificial intelligence models — including Anthropic’s Claude and Alibaba’s Qwen — to mass-generate pro-Russian articles and manipulate other chatbots.

Leaked files from the laptop of former US Marine turned Russian propagandist John Mark Dugan have revealed the extent of his disinformation work. The documents show that he systematically used artificial intelligence models — including Anthropic’s Claude and Alibaba’s Qwen — to mass-generate pro-Russian articles and manipulate other chatbots.

How the AI-based content “factory” worked

Data on Dugan's working methods appeared online after unknown people hacked his account on the X social network and published a link to a duplicate site with his emails, correspondence on Telegram and Signal, as well as a history of requests (prompts) to AI, writes Cybernews. Dugan himself confirmed the authenticity of the documents in a to NewsGuard, stating that he did not care the publication of the contents of his laptop.

According to the leak, the main goal of the campaign was the so-called “infection” of large language models (LLMs) and algorithms:

The network created up to 43,000 articles per month that praised the Russian Federation and disguised themselves as legitimate materials for a Western audience.

The network created up to 43,000 articles per month that praised the Russian Federation and disguised themselves as legitimate materials for a Western audience.

articles were constantly updated with AI-generated text to increase their rankings in engines and make them available as “sources” for other chatbots;

articles were constantly updated with AI-generated text to increase their rankings in engines and make them available as “sources” for other chatbots;

The cost of this activity was 2 million rubles ( $23,000) per month. In one of his impassioned letters to Claude, Dugan asked the AI ​​to provide an estimate, complaining that the current volume of 43,000 texts was becoming too expensive.

The cost of this activity was 2 million rubles ( $23,000) per month. In one of his impassioned letters to Claude, Dugan asked the AI ​​to provide an estimate, complaining that the current volume of 43,000 texts was becoming too expensive.

Bypassing Anthropic and Alibaba security policies

The propagandist used Anthropic Claude, the Chinese Alibaba Qwen model, and the Open WebUI interface, which allows you to run models on your own terms, to work. Such activity directly violates the terms of use of both platforms, which prohibit the generation of fakes and disinformation.

In his correspondence with Claude, Dugan, in particular, asked to develop the concept of a “freedom of speech site for Europe” aimed at a French-speaking audience, consulting with the bot on the name. And he asked the Qwen model how to write texts so that they could not be recognized by AI content detectors.

Who is John Mark Dugan?

John Mark Dugan (also known by the alias BadWolf) previously served in the US Marine Corps and worked as a deputy sheriff in Florida and Maine, from where he was fired for misconduct and harassment.

In 2016, the FBI opened an investigation into him over the leak of confidential police records. After being indicted on 21 criminal charges in Florida, Dugan fled to Russia, where he was granted political asylum and Russian citizenship in 2023. According to investigators, he is working with the Russian military intelligence agency (GRU), which funds his network of fake websites disguised as American media outlets.