LLMs can create 10,000 malware variants evading detection with 88% success, degrading ML classifiers and risking AI model ...