A study finds that large language models struggle to interpret negation in figurative language. Performance varies significantly based on prompt style, highlighting a key limitation in current models' understanding of complex linguistic structures.