New OpenAI Aardvark agent automates vulnerability management

CyberScoop reports that OpenAI has unveiled in beta its new Aardvark security-focused artificial intelligence model that seeks to advance vulnerability discovery, patching, and remediation efforts.

Aardvark, which uses ChatGPT-5, scans repositories not only to identify, prioritize, and address vulnerabilities but also to create threat models tailored to security goals, according to OpenAI, which touted the new system's 92% accuracy in determining known and synthetic flaws in golden repositories.

"By catching vulnerabilities early, validating real-world exploitability, and offering clear fixes, Aardvark can strengthen security without slowing innovation," said OpenAI, which noted expanded tool availability after the implementation of further vulnerability detection, validation, and reporting capability enhancements.

Such a development comes months after Oege de Moor, founder of AI penetration testing startup XBOW, disclosed the mostly automated operations of the firm's bug-hunting model. However, XBOW's compute costs were noted by de Moor to be inadequately supported by bounties the company has received for addressing thousands of software flaws, de Moor said.