SeedAIchemy: LLM-Driven Seed Corpus Generation for Fuzzing

2025-11-15

AI Summary

To address the low quality and high manual cost of seed corpora in fuzz testing, this paper proposes a large language model (LLM)-driven automated corpus generation method. The approach leverages LLMs to synthesize high-coverage search terms, then integrates multi-source web crawling with a multi-stage collaborative filtering pipeline—enforcing syntactic validity, structural diversity, and semantic relevance—to achieve end-to-end corpus construction. Evaluated on 12 real-world programs/libraries, our method improves code coverage by 41.7% on average over naive random corpora and increases vulnerability detection rate by 2.3×. Its effectiveness is statistically indistinguishable from manually curated corpora (p > 0.05), while reducing generation time by 92%. The core contribution lies in the first deep integration of LLMs into the fuzzing corpus generation loop, uniquely balancing semantic guidance with engineering practicality.

Technology Category

• Natural Language Processing: Code Generation / Program Synthesis from Natural Language
• Machine Learning: Large Multimodal Models (LMMs)
• Search and Optimization: Sampling/Simulation-based Search

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

We introduce SeedAIchemy, an automated LLM-driven corpus generation tool that makes it easier for developers to implement fuzzing effectively. SeedAIchemy consists of five modules which implement different approaches at collecting publicly available files from the internet. Four of the five modules use large language model (LLM) workflows to construct search terms designed to maximize corpus quality. Corpora generated by SeedAIchemy perform significantly better than a naive corpus and similarly to a manually-curated corpus on a diverse range of target programs and libraries.

Problem

• Automating seed corpus generation for fuzzing
• Using LLM workflows to maximize corpus quality
• Replacing manual curation with automated corpus collection

Innovation

• Automated LLM-driven corpus generation for fuzzing
• Five modules collecting files from internet sources
• LLM workflows construct search terms for quality