Disinformation and deepfakes are known risks but biased LLMs could also be used to influence public opinion