Marina Lyubimova
● OpenAI just unveiled Aardvark—an autonomous security researcher running on GPT-5 that finds, evaluates, and patches vulnerabilities on its own. As Tibor Blaho explained, Aardvark constantly monitors Git repositories, checks how exploitable issues are, ranks them by severity, and generates fixes using AI reasoning and Codex—skipping traditional scanning methods entirely.
● The system works in stages: it scans code for vulnerabilities based on an internal threat model, tests each one in a sandbox, then uses Codex to create patches. After human review, it submits them as automated pull requests.
● In testing, Aardvark hit 92% accuracy detecting both known bugs and synthetic ones in curated repositories. It's already found real vulnerabilities in open-source projects—ten have gotten CVE identifiers. Tibor Blaho noted that "Aardvark" first quietly appeared in ChatGPT around September 12, 2025, but stayed under wraps until now.
● Right now, it's only available to internal teams and select testers through a dashboard in ChatGPT, where users can schedule scans, connect repos, and manage findings.
Marina Lyubimova
