OpenAIPatent applicationSecurity & privacyInterest 7/10

OpenAI patents AI agents that find security flaws in code

Patent drawing
Patent drawing from the official filing.

This filing describes a system that uses large language models to scan computer code for security weaknesses. It ranks functions by how they connect to other parts of the app and then generates specific threat scenarios to identify potential vulnerabilities.

Why it mattersDevelopers and security teams could use this to automatically spot risks in software, potentially making apps harder to hack.

In the filing’s own words

The present disclosure relates to systems, non-transitory computer-readable media, and methods for analyzing vulnerabilities within a computer application. The disclosed systems can retrieve, utilizing a code indexing subroutine, a computer function from a function repository of a computer application. Moreover, the disclosed systems can determine a vulnerability ranking for the computer function based on a call graph. The disclosed systems can generate, utilizing a large language model, a threat model comprising a structured data object defining threat parameters corresponding to the computer application. Additionally, the disclosed systems can generate threat objects based on vulnerability rankings and utilize a large language model to analyze computer functions in conjunction with the threat model, thereby indicating threat data associated with the computer function.

What exactly they claim (claim 1)

1. A computer-implemented method comprising: retrieving, utilizing a code indexing subroutine, a computer function from a function repository of a computer application; determining a vulnerability ranking for the computer function based on a call graph defining relationships among computer functions of the computer application; generating, utilizing a large language model of a generative artificial intelligence (AI) system, a threat model comprising a structured data object defining threat parameters corresponding to the computer application; and generating, based on the vulnerability ranking and by utilizing a large language model to analyze the computer function together with the threat model, a threat object indicating threat data associated with the computer function.

Facts

Company
OpenAI (OpenAI OpCo, LLC)
Official title
DETECTING VULNERABILITIES IN COMPUTER FUNCTIONS UTILIZING GENERATIVE MODEL AGENTS
Publication number
US20260350180A1
Application number
19365022
Filed
Oct 21, 2025
Published
Nov 5, 2026
Inventors
3
Claims
20

Classified under

G06F 21/577 Physics & computingG06F 21/53 Physics & computingG06F 2221/034 Physics & computing

More from OpenAI in Security & privacy

Summaries are simplified from the filing’s own abstract and first claim. A patent filing is not a product announcement.
Patent data and drawings: U.S. Patent and Trademark Office (USPTO). Nebula is not affiliated with or endorsed by the USPTO.