helpnetsecurity.com·17 november 2026

Too many instructions can hold back AI security audits - Help Net Security

Image for the article: Too many instructions can hold back AI security audits - Help Net Security
Image via helpnetsecurity.com

GitHub uses AI agents to audit software for vulnerabilities, focusing initially on authentication and authorization bugs. The process involves splitting audits into small tasks, testing and refining prompts, and producing threat models that list components and entry points. Every report is manually validated and reviewed before being sent to maintainers, ensuring a 95% fix rate. GitHub also works to reduce the cost of false positives for maintainers by improving the reporting experience and introducing features like rate limits and structured forms.