"While all previous large language models had a near zero success rate in actually executing exploits, Mythos successfully breaches targets 72.4% of the time."