// DARPA AIxCC FINALIST · OPEN SOURCE

From zero-day to merged patch, autonomously.

FuzzingBrain is a Cyber Reasoning System that finds, proves, and fixes security bugs without humans in the loop — 124 zero-day vulnerabilities reported across 53 open source projects, 82 already patched upstream.

View Public Disclosures Source Code Leaderboard
124 // vulns_reported
53 // projects_targeted
82 // patches_merged

// 01 — SYSTEM

What FuzzingBrain does

An end-to-end pipeline that reasons about code, proves vulnerabilities, and ships fixes — autonomously.

Autonomous Detection

Automatically generates Proofs-of-Vulnerability (PoVs) and produces patches for discovered security issues — no human intervention required.

LLM-Powered

Leverages 23 distinct LLM-based strategies across frontier models from Anthropic, Google, and OpenAI for comprehensive analysis.

Massively Parallel

Deployed across ~100 VMs with thousands of concurrent threads, enabling rapid vulnerability discovery and patch generation.

// 02 — ARCHITECTURE

Technical approach

Four services, running in parallel

  • CRS Web Service Central coordinator for task decomposition and fuzzer distribution
  • Static Analysis Function metadata, reachability, and call-path analysis
  • Worker Services Parallel PoV generation and patching strategies
  • Submission Service Deduplication, SARIF validation, and bundling

PoV Generation

delta-scan full-scan sarif-based

10 LLM-based strategies for vulnerability discovery, from iterative refinement to multi-input generation with coverage feedback.

Patching

multi-model xpatch path-aware

13 patching strategies including our novel XPatch approach that generates patches even without a PoV.

Key technical innovations

Iterative LLM Refinement

Multi-turn dialogue with structured feedback loops incorporating execution results and coverage data.

Multi-Model Fallback

Resilient architecture with automatic model switching when individual LLMs fail or reach limits.

Static / Dynamic Integration

Call paths, reachability, and real-time coverage feedback guide vulnerability discovery.

// 03 — RESULTS

Proven in competition and in the wild

124
Reported Upstream
Across 53 open source projects
82
Patches Merged
Fixed in upstream releases
<5%
False Positive Rate
4-principle verification
// affected_projects
CUPS Ghidra OpenLDAP Apache Avro ImageMagick simdutf UPX BlueZ
View all 74 public disclosures Read the full report

Performance insights

Speed

Sub-5-minute first findings on multiple targets.

Effectiveness

AI-generated harnesses found bugs in 26 of 26 targeted projects.

Scalability

Scaled from competition VMs to continuous open source fuzzing.

// 04 — TEAM

Research team

Ze Sheng

Texas A&M University

Qingxiao Xu

Texas A&M University

Zhicheng Chen

Texas A&M University

Jianwei Huang

Texas A&M University

Matthew Woodcock

Texas A&M University

Heqing Huang

City University of Hong Kong

Alastair F. Donaldson

Imperial College London

Guofei Gu

Texas A&M University

Jeff Huang

Team Lead

Texas A&M University

// 05 — RESOURCES

Open source & resources

FuzzingBrain CRS

Complete Cyber Reasoning System with all 23 LLM-based strategies.

View on GitHub

LLM Leaderboard

Benchmark comparing state-of-the-art LLMs on vulnerability detection and patching.

View Leaderboard

Technical Paper

Detailed description of our CRS with emphasis on LLM-powered components.

Read Paper