Cloudflare says OpenAI models aid vulnerability discovery and fixes
News
Cloudflare says it is using OpenAI Daybreak models to investigate vulnerabilities and propose fixes, with Codex supporting developer productivity.
Cloudflare says it is using OpenAI Daybreak models to investigate vulnerabilities and propose fixes for review, adding another layer of AI-assisted security tooling to its development workflow. The company also said its developers use Codex to build faster and improve code quality, according to material published by OpenAI on 2026-10-21.
Evidence and context
The OpenAI post states that Cloudflare links OpenAI Daybreak to vulnerability discovery and defense, with the models used to investigate weaknesses and propose remediation steps for human review. The disclosure is consistent with Cloudflare’s earlier 2026-09-03 announcements, which described a Vulnerability Discovery and Remediation service as an invitation-only early-access offering through Cloudflare Managed Defense for selected Enterprise customers.
That earlier Cloudflare material said the workflow combines OpenAI Daybreak models, including GPT-5.6 Cyber, with Cloudflare production-traffic and security context to investigate customer-authorized code, prioritize vulnerabilities, validate findings and propose code patches. It also said the system can suggest custom web application firewall rules where evidence supports them. The company specified that model inference runs on OpenAI servers and that prompts are routed through Cloudflare AI Gateway, while proposed changes still undergo checks and human review.
The available evidence does not establish general availability, and the background material says the service remained invitation-only early access as of the 2026-09-03 announcement. OpenAI’s Daybreak partner page also described the Cloudflare relationship as part of a controlled rollout, with interest registration not guaranteeing inclusion, model access or a production timeline.
