Cloudflare links vulnerability discovery to defense with OpenAI

Cloudflare has integrated OpenAI models accessed through Daybreak into its Vulnerability Discovery and Remediation service, an early-access offering under Cloudflare Managed Defense. The models analyze customer-authorized code to identify potential vulnerabilities, leveraging live traffic data and security signals to prioritize findings. Proposed mitigations, including code patches and custom firewall rules, are prepared for customer review, with automated checks and team validation ensuring accuracy before implementation.

The service aims to address the challenge faced by security teams overwhelmed by the volume of vulnerability alerts and data. By connecting source code vulnerabilities to production activity, Cloudflare provides context to help teams prioritize investigations. Security teams receive evidence-based findings, proposed protections, and code fixes, enabling faster decision-making while maintaining control over changes to their applications.

Cloudflare also employs OpenAI’s Codex to enhance its engineering processes, improving developer productivity and code quality. Anderson highlights a recent project where a team of four developers rebuilt a system in three days using Codex, estimating it would have required months and ten times the team size without the tool. The project cost $1,100 in tokens, demonstrating significant efficiency gains.

The integration of OpenAI models into Cloudflare’s workflow reflects a broader shift toward embedding AI into engineering processes. Anderson emphasizes the importance of treating AI as an integral part of the build process rather than a separate tool. This approach aligns with Cloudflare’s goal of providing security teams with actionable insights to prioritize and mitigate vulnerabilities while maintaining customer control over implementation.