The AI Overlords are coming…
Large Language Models (LLMs) demonstrate increasing proficiency in handling complex tasks, Agent technology is emerging as a new paradigm in the field of vulnerability discovery. Since Google Project Zero released Naptime there has a been a increase in the ability of LLM’s and AI models for the use of vulnerablity research.
Notes
That type of reasoning loop could be translated into prompting and context, sort of a realtime version of RAG, and fed into a model that maybe had been trained or at least finetuned on code and security specifics.
Structure
Apparently LLM’s and Agents work better using xml then JSON since they are trained on the entire XHTML of the internet