OpenAI: Codex Security Tracks Software Vulnerabilities

OpenAI recently introduced Codex Security, an AI-based security agent designed to identify vulnerabilities in software projects. This tool, which was previously known by the code name "Aardvark," is now available in research version for users of ChatGPT Enterprise, Business, and Edu. For the first month, access to this tool is free.

Codex Security works by analyzing code repositories, building a threat model specific to each project, and testing the identified vulnerabilities in secure testing environments. According to OpenAI, during its beta phase, the system successfully reduced false positives by over 50% and decreased redundant alerts by 84% in one particular case.

Over the last 30 days, Codex Security has reviewed more than 1.2 million commits and reported 792 critical vulnerabilities. OpenAI has also detected flaws in several open-source projects, including OpenSSH, GnuTLS, GOGS, Thorium, and Chromium, with 14 CVEs published to date. A program aimed at supporting maintainers of open-source projects is currently in development.

Meanwhile, Anthropic has launched its own cybersecurity tool, leading to a decline in stocks within the cybersecurity sector. These developments indicate increasing competition and rapid evolution in the field of software security.