Research: defenders using generative AI to simulate malware variants before they exist in the wild

Came across this paper and thought it was worth sharing here since it addresses something that comes up a lot in ML-based detection discussions. The core problem it tackles: you cannot train a good detector for malware