A new study from Estonia demonstrates that many AI systems can still be manipulated by propaganda-style prompts, with varying degrees of vulnerability depending on the specific model and language used.