Recent research reveals that large language models consistently uphold false beliefs even when presented with explicit denials or corrective warnings. This “negation neglect” suggests that model alignment and truthfulness remain significant challenges for developers.