Cloudflare has launched early access to Vulnerability Discovery and Remediation, a service integrated into Cloudflare Managed Defense. The system utilizes OpenAI's GPT-5.6-Cyber model via the Daybreak Defense Network to identify software flaws in customer applications and block attacks at the network edge.
The process begins with a snapshot of Web Assets and firewall events, followed by a reconnaissance agent mapping request paths to the codebase. Hunter agents then analyze the map, including code from Cloudflare Workers. Findings are validated and risk-rated based on production traffic and active probing evidence.
The service provides two types of fixes: custom firewall rules as temporary stopgaps and drafted code patches for engineer review. Both require explicit human approval before implementation, as the models cannot apply changes autonomously. All prompts pass through Cloudflare AI Gateway, with redaction used to remove unnecessary context.
The need for the service is highlighted by National Vulnerability Database statistics: 60,475 vulnerabilities were recorded by September, compared to 48,185 for the entire year of 2025. The system aims to filter these thousands of findings to identify those critical to production.
GPT-5.6-Cyber, released in August, is part of the Daybreak Red tier. It achieved a 95% success rate on OpenAI's advanced cybersecurity benchmark, significantly higher than the 1.5% rate of the general-purpose GPT-5.6 Sol model.
Other industry players are also adopting these models: Palo Alto Networks integrated them into Unit 42 consulting, and Proofpoint added them to its SOC Analyst Agent. Additionally, OpenAI committed $1 billion to subsidize Daybreak access for utilities, local governments, community banks, and nonprofits.
