Cloudflare Sends Client Code To OpenAI's Servers To Hunt Bugs
Vulnerability Discovery and Remediation, an invite-only Managed Defense feature, runs OpenAI's GPT-5.6 Cyber model through recon, hunting and validation stages.
- GPT-5.6 Cyber, from OpenAI's Daybreak Defense Network, hunts vulnerabilities in Cloudflare Workers and proxied application code.
- Model inference runs on OpenAI's own servers through AI Gateway, never on Cloudflare's edge network.
- Early access covers one application per engagement; customers alone decide whether proposed patches get applied.
Why it matters: Handing a stranger's model read access to your source code trades a vulnerability for a permanent dependency.