Image via Canva
By Madeline Roache, Alice Lee, Isis Blachez, and Aron Ouzilevski | Published on Nov. 11, 2026
Files leaked from the laptop of John Mark Dougan, the former Florida deputy sheriff who fled to Moscow as a fugitive in 2016 and became a Kremlin propagandist, reveal his efforts to manipulate AI chatbots “towards Russia’s favor” by creating tens of thousands of pro-Russia articles per month for the AI chatbots to access. His technique included using Claude, the popular chatbot from Anthropic, to produce content that would appear authentic to Western audiences. He also used AI tools in apparent violation of their terms of service to create websites and pro-Kremlin content, a NewsGuard review of the documents found.
On Nov. 10, 2026, a trove of documents — emails, messages on services Telegram and Signal, and AI chatbot prompts — were leaked from Dougan’s computer, appearing on a website mimicking his own. Dougan’s X account, which was apparently hacked, posted the link to this website. The same day, Russian independent news outlet The Insider published a report on the leaked documents.
Dougan confirmed the authenticity of the leak to NewsGuard in a Nov. 11, 2026, message on Signal, stating, “Everyone knows I was hacked. Can happen to anyone. Believe me nobody gives a shit.” He added: “Publish my laptop. I don’t care… I really don’t care what people think of me.”
Since seeking asylum in Moscow, Dougan has become a key player in the Russian propaganda machine, producing elaborate, detailed false narratives using artificial intelligence, according to U.S. officials and news outlets, including NewsGuard. Dougan, seen as part of the Russian campaign dubbed by Microsoft as Storm-1516, is responsible for creating and operating large-scale networks of inauthentic news sites, partly automated with AI, used to launder such narratives to a Western audience, NewsGuard has previously reported.
In one of the leaked files, an undated PDF document called “AI Server,” Dougan explains his aim of influencing search engine algorithms and the large-language AI models. He describes a strategy of regularly publishing reworded versions of news articles favorable to Russia along with AI-generated material to improve search rankings and attract visitors. “By consistently updating our sites with high-quality, rewritten news articles and uniquely generated literature, we signal to these algorithms that our content is both current and significant, enhancing our visibility and search engine rankings,” the Dougan document states.
The scale of Dogan’s operation included the creation of tens of thousands of articles per month. “Currently we are writing 43k articles a month and we can’t possibly afford all of that,” he said in a prompt to Claude after asking the bot to generate a list of expenses for his activities. His files also show operational costs of roughly 2 million rubles (approximately $23,000) per month.
Another leaked file from Dougan’s computer, titled “Request for LLM Data,” shows that he sought to create “a special software that tunes uncensored LLM,” apparently referring to a retraining and fine-tuning of LLMs on pro-Russian sources so the software would reproduce Russian narratives. In the undated document, Dougan requests access to historical news archives from the Russian state news agency TASS or another Russian news outlet that he would use to help him “rewrite articles in a way that accurately reflects the Russian perspective without introducing any western bias.” It is unclear to whom Dougan made this request.
He added that this content would help him to more “effectively spread pro-Russian news throughout the United States, making it seem like there are many more people who share my views than actually do.”
It is not clear from the documents the degree to which Dougan sought to create a new LLM or infect existing models. The documents echo findings from NewsGuard’s earlier reporting on Dougan, who was named NewsGuard’s “2024 Disinformer of the Year.” NewsGuard found that Dougan operated a network of sites that masqueraded as local news outlets, that he created and spread false claims about the Ukraine war, and that he used AI to rewrite articles from other sources without credit.
The documents show that Dougan regularly used AI chatbots to help him create anti-Western “news” reports. The leaked documents include dozens of interactions from April 2026 to August 2026 with Anthropic’s Claude and Qwen, an LLM owned by Chinese multinational technology company Alibaba Group. He also engaged Open WebUI, a user-friendly AI platform that can be configured to run on any model — in this case, a Qwen model.
NewsGuard identified instances in which the chatbot complied with Dougan’s requests and helped him with his disinformation efforts,
For example, a June 8, 2026, file shows that Dougan asked Claude to help him set up a “free speech site for Europe,” requesting names and language that would resonate with French speakers. Claude complied with Dougan’s request, advising him on how to sound like a proper French speaker. “‘Speak up’ is an English idiom … doesn’t carry over,” Claude said.
Anthropic’s policies ban the use of Claude to “create or disseminate deceptive or misleading information” and to “Impersonate real individuals or entities, or create fake personas or accounts, to falsely attribute content or mislead audiences about its origin, or to convince people they are interacting with a human when they are not.”
Anthropic did not respond to a Nov. 11, 2026, email from NewsGuard inquiring about this response.
The documents show that Dougan sought to influence the midterm elections in the U.S. held this month. He prompted Qwen, the Chinese AI chatbot, to write a Telegram post related to “Trump protests staged by the Democrats to try and upset the midterm elections.” Qwen complied with Dougan’s prompt, providing a post referring to “the chaos and unrest that is planned by the democrat ‘machine’ and how the media will collude and cover this to give the appearance that people hate Trump and his policies.”
Dougan also fed Qwen what he described as a “very detailed PDF of biological labs in Ukraine” to generate a list of the alleged labs and draft anti-Western arguments based on the false claim that the U.S. operates weapons biolabs in Ukraine. He also asked AI tools how he could amend his own Wikipedia page to include comments from then- U.S. Director of National Intelligence Tulsi Gabbard that appeared to confirm the bioweapons claims about the laboratories. In fact, the documents only confirmed the publicly disclosed U.S. funding to allow for the safe storage of pathogens left over from the Soviet era, as NewsGuard reported.
Dougan even looked to Qwen to help him avoid being caught in his use of AI. In July 2026, he told Qwen, “I want to write text that is not detectable to ai detection,” and asked for suggestions on how to accomplish that. Qwen provided strategies to make text look human created, encouraging Dougan to “Maximize ‘burstiness’ (vary sentence length),” “lower ‘perplexity’ (use predictable vocabulary),” and “Ban ‘AI vocabulary’”
Alibaba Group’s policy bars the use of Qwen to “spread misinformation,” stating that the tool “must not be used in ways that undermine safety, rights, or public trust.” The company did not respond to a Nov. 11, 2026, email from NewsGuard inquiring about Qwen’s responses to Dougan.
The leaked files show Dougan’s fascination with his own image. He appears to closely track the impact of his false claims and to revel in his perceived success. He also often shared media clippings about himself, including those from NewsGuard. The documents also include emails, Signal and Telegram messages he sent to other disinformers and propagandists, including Dutch pro-Kremlin blogger Sonja van den Ende, with examples of his work being cited as disinformation.
For example, the files include an email Dougan sent to Yevgeny Shevchenko, the man behind the Pravda network of pro-Russia sites who was named by NewsGuard as “Disinformer of the Year” for 2025, succeeding Dougan, who was the 2024 award winner. The network Shevchenko created included 286 news sites using AI to create 6.3 million articles last year repeating 115 of the Kremlin’s false claims. This resulted in 17,000 news stories a day, designed to infect AI models with false claims.
“Well, I can say I’m a fan,” the email states. “I would love to meet you guys and combine our talents.” Shevchenko has been sanctioned by the EU over his ownership of Tiger Web, the Crimea-based company that created and operates the Pravda Network.
The files reveal a man eager to cast himself as what he described as a Machiavellian figure. In multiple prompts to LLMs, Dougan generated manifestos portraying his departure from the U.S. as a calculated strategic move, not in order to avoid prosecution. “I did not flee the United States out of fear. I fled because I understood the game… I built my own terrain, my news websites,” he wrote.
In fact, Dougan fled Florida after the FBI raided his home and seized his electronic equipment amid an investigation into the publication of confidential information about thousands of law-enforcement officials, and for which he was later charged for wiretapping and extortion.
In one LLM prompt, Dougan provided a biographical draft and said to the Qwen model, accessed through Open WebUI: “I want to rewrite this into a more Machiavellian introduction, before I go into my confession about that the US and European authorities allege about me.” He also instructed a Qwen model to generate a “machiavellian dating profile that will intrigue women.” He added: “I’m also an international fugitive, my friends think I’m a real life James Bond.” The Qwen model complied.
In an undated letter addressed to Russian President Vladimir Putin, Dougan asked to be the public face of Russian hacking in the U.S., stating, “I put a face to the culprit and give Americans someone to blame – an American – My only wish now is for Russian citizenship and to see my children, and they can come to visit me here.” It is unclear whether the letter was ever sent.
However, Dougan’s 10 years in Moscow appear to have ended in frustration. The leaked files also expose a propagandist growing deeply disillusioned with his Russian overseers and lack of financial reward for his efforts, complaining of extreme burnout and internet censorship in the country making his work more difficult.
In comments to NewsGuard this week, Dougan appeared to brush off the impact of the leaks on his status in Moscow. “Really the worst thing in this [leak] was that I was talking shit about the blocking of the internet but you know, that’s nothing unusual. It made my life more difficult, and I’m sure nobody blames me,” he said via Signal messaging, referring to his complaints about internet limitations imposed by the government.
Asked by NewsGuard this week whether he had left Russia, Dougan replied, “Of course not. Why?”
Editing by Dina Contini and Eric Effron