Vulnhuntr is an open-source security tool from Protect AI that uses large language models to discover vulnerabilities in Python codebases with zero-shot analysis. The GitHub project has gathered around 2,800 stars and was last actively updated in 2025. Instead of relying only on fixed rules like traditional static analyzers, Vulnhuntr asks an LLM to reason about data flow through your code, and it made headlines for finding real zero-day vulnerabilities in popular open-source projects.
The tool combines structured initial analysis with iterative, chain-of-thought follow-up: it first maps the codebase and identifies user-input entry points, then prompts the model to trace how that input flows into sensitive sinks such as file operations, SQL queries, and template rendering. It focuses on vulnerability classes LLMs can reason about well, including local file inclusion, server-side request forgery, SQL injection, cross-site scripting, and insecure direct object references, and it produces a report with the vulnerable code and its reasoning.
Vulnhuntr’s authors documented the zero-days it uncovered in projects such as LangChain-related tooling and other widely used Python packages, which helped establish LLM-assisted auditing as a credible complement to fuzzing and rule-based scanning. The tool analyzes a target repository locally, so proprietary code is only shared with the model provider you configure. Its focused scope on Python web and AI application code keeps the analysis depth high where it matters most.
Vulnhuntr is free and open source under the AGPL-3.0 license. You only pay for the LLM API calls it makes during analysis.
Security researchers, AppSec teams, and open-source maintainers who want an AI-assisted second pass over Python code to catch logic-level vulnerabilities that rule-based scanners miss.