LLMs at the Forefront: Pioneering the Future of Fuzz Testing
Introduces FuzzForest, an open-source framework leveraging LLMs to automatically generate, repair, and triage fuzz test harnesses across Python libraries.
Summary
Integrating Large Language Models into Fuzzing Workflows
While Large Language Models (LLMs) often struggle with whole-system architectural reasoning, their ability to synthesize syntactically valid code makes them exceptionally suited for generating fuzz test harnesses.
Overcoming Obstacles in Automated Harness Generation
Human security researchers spend significant time writing initial fuzz harnesses and mocking complex inputs. LLMs can automate harness construction, input synthesis, and triage analysis.
Introducing FuzzForest for Python Libraries
The author introduces FuzzForest, an open-source tool that prompts LLMs to generate and iteratively repair Python fuzzers based on runtime crash feedback.
Discovering Dozens of Bugs in Top Open-Source Projects
Deploying FuzzForest across the 20 most popular open-source Python repositories uncovered dozens of previously undiscovered memory-safety, logic, and parsing bugs.
Related talks
Briefings·24 min
Qinrun Dai · Yifei Xie
Vulnerability discovery traditionally relies on two primary approaches: manual auditing and fuzzing.
Main Stage·42 min
DARPA AIxCC Finalists
Finalists from DARPA's AI Cyber Challenge present autonomous cyber reasoning systems designed to find, patch, and remediate critical software vulnerabilities.
Main Stage·50 min
Olivier Bilodeau · Estelle Ruellan
Demonstrates leveraging multimodal LLMs to analyze millions of infostealer infection screenshots and autonomously extract indicators of compromise.