Malware authors have long used polymorphic techniques to generate variants within a malware family, complicating detection. A new arXiv preprint, arXiv:2605.03619v3, asks whether large language models turn this into an "infinite mutation engine" by automatically producing structurally diverse offensive payloads.
The abstract argues that generative AI enables attackers to synthesize such payloads without manual effort, potentially scaling up the kind of polymorphism traditionally seen in malware families. The title's question is central: whether LLMs can produce effectively unlimited mutation, and what that means for defenses.
Because the available source is only the abstract, the article does not report specific experimental results or metrics. This is a single-source item, so there are no conflicting findings to compare.