NEW REPORT: Russian propaganda may be flooding AI models

Click here to read the report.

Click here to read the press release.

Click here to access the database.

It may be tempting to consider Russian influence campaigns in the United States and other democratic countries as merely an afterthought, given the chaos and uncertainty that Donald Trump’s second presidency is causing. But his reelection–and the growing influence of Russia-friendly, far right parties worldwide–only makes these campaigns more relevant. The Trump administration is actively undoing protections against foreign kleptocratic dealings and disinformation in the United States while the Trump-led United States is apparently siding with authoritarian Russia against democratic states like Ukraine. These actions are essentially putting a for-sale sign on the American information space and dealing a devastating blow to democratic and information resilience worldwide.

The chaos of the second Trump administration also serves as a distraction that allows Russian influence campaigns to continue to grow unchecked. The American Sunlight Project’s latest report explores the newest expansion of the so-called “Pravda network,” a collection of web pages and social media accounts that aggregate pro-Russia propaganda around the world. This expansion explicitly targets the United States, among dozens of other new targets that include countries in Africa, Asia, Europe, and North America. The network is also newly targeting prominent heads of state, international organizations, and commonly spoken languages.

More concerning than the network’s geographic expansion are its growing capabilities and how other information operations could copy the Pravda network’s model in the future. The network itself is a highly centralized system of largely automated propaganda aggregation, translation, and sharing across the network’s massive footprint. ASP estimates that the network is producing at least 3 million articles of pro-Russia propaganda per year, a number that does not include the network’s activity on X (Twitter), Telegram, Bluesky, and VK. Given the past growth of this network, this rate is likely to increase.

Such a large network poses several risks. Both the general public and prominent leaders are more likely to see the network’s content–which is laden with harmful, anti-democratic disinformation–given how pervasive it is. This means that audiences might be more likely to believe and share this content, and the content is more likely to be wittingly or unwittingly laundered by more legitimate sources to even bigger audiences. Pravda network articles have already been cited on Wikipedia, for example, and Trump’s recent accusation that Ukraine was somehow responsible for the war clearly shows that influential figures in U.S. politics readily repeat Russian disinformation.

The Pravda network also appears to be set up to flood large-language models with pro-Kremlin content. Past research shows that all ten major AI chatbots can reproduce Russian disinformation in response to certain prompts, suggesting that their training data contained Russian disinformation to begin with. The Pravda network is rapidly churning out content without clear intention of building actual human readership on its poorly designed websites means that it may be aiming to land its content in LLM training data. Without more robust safeguards on training datasets and LLMs alike, the risk that the network’s content is eventually parroted by AI chatbots and other LLMs is very large–regardless of the network’s intent.

Thus, we enter a new chapter of Russian information operations. Democracies not only have to contend with a chaotic U.S. government on the backdrop of ever-flourishing Russian disinformation; advancements in technology such as AI also promise to turbocharge foreign influence campaigns in unforeseen ways. Major players in government, civil society, and the private sector must act to ensure transparency and oversight of AI training models and prevent the most harmful effects of this new era of information operations, and the public should understand that the internet they are navigating becomes more misleading by the minute.