The Infinite Mutation Engine: Measuring Polymorphism in LLM-Generated Offensive Code

arXiv·medium signalXBlueskyLinkedInCopy link

Researchers measure how effectively LLMs can serve as polymorphic malware engines — generating structurally diverse payloads with identical behavior that evade signature-based detection. Integrating generative AI into offensive toolchains enables attackers to synthesize variants at scale, challenging traditional AV/EDR approaches. The quantitative measurements of polymorphic diversity provide defenders with concrete data on what detection methods still work and which ones LLM-generated variants reliably bypass.