Project
LLMForFuzzing
Prompt engineering and Python automation that helped language models participate in vulnerability discovery and contributed to a real zero-day finding.
Python · ChatGPT API · Prompt Engineering · Fuzzing · Cybersecurity
Product Snapshot
- Role
- Core contributor
- Focus
- PoC reproduction, natural-language semantic mutation, Adobe/Foxit JavaScript APIs, and automated test-PDF generation.
- Validation
- Project records report 24/32 reproduced PoCs, 200+ APIs, roughly 2,000-3,000 test PDFs, and one UI-related zero-day contribution.
- Public Proof
- Source, test samples, and closeout materials; scale and zero-day results were not independently rerun for this portfolio.
Next Step
Starting point
LLMForFuzzing was one of the earliest AI × security projects I joined.
The hard part was not calling an API. It was finding ways to make model outputs concrete enough to be useful inside a real vulnerability-discovery process.
What I worked on
- prompt engineering for plausible exploit paths
- Python automation that reduced repetitive manual trial and error
Why it mattered
Project records report 24/32 reproduced PoCs, 200+ Adobe/Foxit JavaScript APIs, roughly 2,000-3,000 generated test PDFs, and one UI-related zero-day contribution.
Source, test samples, and closeout materials corroborate the workflow. The scale and zero-day result come from project records; the original fuzzing run was not repeated for this portfolio.
Configure the public Giscus environment variables to open discussion on the public site.