Malware That Thinks: Polymorphic and Metamorphic Malware Using AI Agents

Traditional malware is a brainless beast. It relies on static signatures, pre-packaged decryption loops, or hard-coded command lists that modern Endpoint Detection and Response (EDR) agents easily identify using heuristic API tracing and behavioral call graphing. If the bytes don't change, the hash is burned; if the instructions change mechanically, behavioral engines still catch the motif.

But what happens when malware stops executing static code and begins to think?

In this talk, we shatter the industry’s comfort zone by deep-diving into Generative Polymorphism—a paradigm shift where malware leverages Large Language Models (LLMs) and autonomous AI agents to evaluate its environment, synthesize unique code on-the-fly, and dynamically adapt its attack path.

We will deep dive into the most modern and future evolutions of polymorphic and metamorphic malware - the ones that'll use AI models to change shape and rewrite payload.

We will also take a look at the architecture of AGAT - how to build a research and analysis tool specifcally for AI powered polymorphic malware.