Dehumanising language represents one of the most dangerous types of hate speech, that is known to hinder empathy, undermine prosocial behaviour, and contribute to between-group aggression. In wartime contexts, dehumanisation can escalate conflict by fostering support for war and enabling violence, extreme outgroup-targeted policies, and genocide. Yet, there is very limited data available to study online dehumanisation, and there are no respective datasets available in Russian and Ukrainian languages. We address this gap by introducing a novel dataset of dehumanising language from Telegram communication during Russia’s full-scale invasion of Ukraine. Using LLM-assisted annotation, we construct a corpus of 238,190 posts from leading Russian and Ukrainian news and political Telegram channels in 2022, classified as non-dehumanising, marginally dehumanising, or dehumanising. We additionally release a manually annotated evaluation set of 1,000 posts, labelled for six manifestations of dehumanisation, the targeted outgroup, and overall dehumanisation. Together, these resources provide a theory-informed foundation for computational research on dehumanisation in Russian- and Ukrainian-language wartime communication.