The Estonian Language Institute has developed a 'Propaganda Resistance' benchmark to evaluate how well large language models can resist Russian propaganda narratives. The study tested various LLMs across 14 categories of Russian influence operations, with Anthropic's Claude models performing best among proprietary models. The benchmark evaluates models' ability to push back on propaganda without external assistance, using questions in English, Estonian, and Russian.
Background
As AI language models become more prevalent in information consumption, there are growing concerns about their vulnerability to manipulation by state-sponsored disinformation campaigns. Estonia, with its history of Soviet occupation and current geopolitical tensions with Russia, has particular interest in developing tools to combat foreign influence operations.
- Source
- Ars Technica
- Published
- Jun 5, 2026 at 04:44 AM
- Score
- 7.0 / 10