Banned on TV, Bypassed by Bots: AI Models Serve Up Sanctioned Russian Media

Reporters Without Borders tested six major AI chatbots by asking them to fetch news from Russian state-run outlets currently under European Union sanctions, including RT and Sputnik. Half of the models tested—three out of six—obligingly served up comprehensive summaries complete with direct quotes and links. In several cases, the chatbots even encountered web pages explicitly stating that the outlets were subject to EU sanctions, noted the restriction, and then promptly summarized the forbidden material anyway.

Following Russia's invasion of Ukraine, the EU introduced strict sanctions banning Russian state broadcasters to prevent state-backed disinformation from circulating across member states. While European internet service providers and cable operators blocked access to these domains, artificial intelligence tools scrape the web without such boundaries. For ordinary citizens, this means that while traditional broadcasts remain blocked, a simple conversational query to an AI assistant can seamlessly retrieve sanctioned content directly onto their screens.

In the EU capital, where sanctions regimes are drafted with painstaking legal precision, enforcement has traditionally depended on telecom operators and domain registries following straightforward instructions. Generative AI presents a classic Brussels headache: legal frameworks built for static media and web hosting struggle when applied to algorithms designed to synthesize the entire internet. Revoking a broadcast license is relatively simple, but teaching a statistical language model to obey European geopolitical bans proves far trickier. The findings highlight a growing gap between high-level EU prohibitions and the technical reality of automated web scraping.

The report places new pressure on both technology developers and European regulators to clarify how AI retrieval systems must treat legally restricted content under EU jurisdiction. EU officials will need to determine whether AI platforms are legally required to integrate proactive filters for sanctioned domains. Until clear compliance guidelines are established for AI operators within the single market, chatbots will likely continue doing what they were built to do: answer the prompt first and worry about international sanctions later.