GitHub Security Lab releases agent workflow for automated fuzz testing

GitHub Security Lab has published a workflow that uses an AI agent to set up and run fuzz tests for C and C++ projects. Given a repository, it selects functions to test, writes test harnesses, runs AFL++, checks which code the tests reach and prepares reports on distinct crashes. The workflow uses coverage results to revise its tests and carries useful inputs into later runs. Its author warns that the agent runs build commands directly on the host, so users should run it in a disposable environment. The generated vulnerability assessments and suggested fixes also require human review.