Work in Progress

Propaganda is already influencing large language models: evidence from training data, audits, and real-world usage