GitHub Launches AI‑Driven Fuzzing Pipeline to Automate Vulnerability Discovery

GitHub’s Security Lab has released the Fuzzing Taskflow, an autonomous fuzzing pipeline that uses a large‑language‑model agent to replace most of the manual work involved in fuzz testing C/C++ projects. The agent automatically analyzes a repository, identifies entry points, writes harnesses, runs AFL++, and iteratively improves coverage by adding seeds, editing harnesses, or enriching dictionaries. Coverage feedback is looped through a SQLite database, and the pipeline stops when two consecutive iterations add less than a configurable 1 % line‑coverage gain. The system also ships structure‑aware mutators for JSON, XML, regex, PNG, and TLV formats. By default it runs on Claude Sonnet 5 and is intended to be executed in disposable environments such as Codespaces or throwaway VMs. The goal is to let developers focus on triage while the agent handles the heavy lifting of fuzzing.