Large-language models are being used intensively by both sides in the Ukraine war, demonstrating their potential for offensive propaganda, but also how far they can help defend against political disinformation. Can AI help democracies win the information war?

As large language models become more and more used in academic and other kinds of writing so model hallucinations risk introducing errors, which can propagate as they become accessed as training data. Without counter-action, this process risks creating feedback loops of misinformation and finally model collapse.