ESTONIAN INSTITUTE TESTS AI VULNERABILITY TO RUSSIAN PROPAGANDA

The benchmark provides researchers with a standardized method to evaluate language models' resistance to propaganda content, particularly Russian disinformation narratives. This comes as AI systems increasingly influence public discourse and decision-making across sectors. The testing framework allows developers and researchers to identify weaknesses in their models before deployment in sensitive applications. Results from such benchmarks can guide improvements in model training and filtering mechanisms. Russian propaganda represents a specific threat vector given documented state-sponsored disinformation campaigns targeting Western countries and Eastern Europe. The Estonian institute's work addresses a gap in AI safety research by focusing on geopolitical vulnerabilities rather than general robustness. Access to standardized benchmarks helps the AI research community understand how easily current models can be manipulated or misled by coordinated disinformation efforts, informing better safety practices.