Researchers from the UK, Germany, and the US discovered a distinct internal representation of pain, termed the "pain axis," across 25 open-weight large language models.