Why AI Gets Crimea Wrong – A Summary of Ivan Dobrovolsky’s Kyiv Independent Article

Photo: Locator map of Krym province.svg, dated November 29, 2010, by Urutseg

This summary is based on Ivan Dobrovolsky’s Kyiv Independent article entitled, Why AI Believes Crimea Is Russian – and what to do about it.

Dobrovolsky argues that leading AI models, including ChatGPT, Claude, and Gemini, often portray Crimea as part of Russia not because of isolated errors, but because they are trained on flawed digital infrastructure and datasets. He traces the problem to widely used open source mapping databases, web content, and even academic publications that incorrectly identify occupied Crimea as Russian.

The article argues that Russian disinformation has become deeply integrated into the digital sources that AI systems rely on, allowing inaccurate information to be repeated and reinforced across widely used AI tools. Dobrovolsky calls on AI developers, open source projects, and academic publishers to ensure their data complies with international law and accurately reflects Ukraine’s internationally recognized borders.

Overall, the article highlights that improving AI accuracy requires correcting the underlying data on which it is trained, not simply improving the AI models themselves. For a deeper analysis, Dobrovolsky’s original article examines how this digital infrastructure became contaminated and what can be done to address it. 

NewsGuard Report: Kremlin-Backed ‘Pravda’ Network Infiltrates AI Chatbots with Russian Disinformation

Photo: Artificial Intelligence, Brain, Think — royalty-free stock image from Pixabay by Gerd Altmann

This summary is from the NewsGuard Report.

A NewsGuard investigation revealed that a Kremlin-backed disinformation network called Pravda has deliberately infiltrated Western generative AI systems with pro-Russian propaganda by flooding the internet with millions of fake news articles designed to manipulate large language models.

The audit of ten leading AI chatbots—including ChatGPT, Gemini, and Copilot—found that they echoed false narratives from the Pravda network 33% of the time, often citing Pravda’s fabricated stories as legitimate sources.

Originating in Crimea and run by the IT firm TigerWeb, Pravda operates across 150 domains in 49 countries, using aggressive search engine optimization (SEO) to dominate online results and “groom” AI models by saturating their data inputs with disinformation.

The campaign, described as “LLM grooming,” seeks to bias AI outputs toward Moscow’s geopolitical agenda, a tactic openly endorsed by former U.S. fugitive turned Russian propagandist John Mark Dougan.

Though the network has little organic human audience, it effectively exploits AI training mechanisms to launder Kremlin propaganda through ostensibly neutral chatbots, posing serious long-term political, social, and technological risks.