Cloudflare is combining OpenAI models accessed through Daybreak with live traffic and security signals from its global network to investigate customer‑authorized source code, identify potential vulnerabilities, and prepare proposed fixes for customer review. This capability is available as an early‑access service called Vulnerability Discovery and Remediation within Cloudflare Managed Defense.
Context and scale
Cloudflare operates one of the world’s largest networks and proxies roughly 20% of the web’s traffic. The company also reported that it blocked cyber threats on average daily across its network in Q1 2026, underscoring the need for faster vulnerability detection and mitigation.
How Vulnerability Discovery and Remediation works
The service uses OpenAI models accessed via Daybreak together with Cloudflare’s production traffic and threat context. The models analyze customer‑authorized source code to flag potential vulnerabilities and assess the evidence supporting each finding.
Cloudflare enriches those findings with live signals — for example, whether the affected route is active, whether it is seeing suspicious traffic, and whether existing security controls are already protecting it. That additional context helps security teams prioritize which issues to investigate first.
When appropriate, the service prepares proposed code patches and custom web application firewall rules to reduce exposure while developers review fixes. Cloudflare runs automated checks on these proposals and its team reviews them before sharing with customers. The result for customers is an end‑to‑end path from detection to protection: evidence to prioritize vulnerabilities, edge protection recommendations to limit exposure, and code fixes to evaluate, while the customer retains control over which changes to implement.
Codex in engineering
Separately, Cloudflare engineers use OpenAI Codex to increase developer productivity, improve code quality, and reduce bugs introduced during development. Anderson notes that traditional security teams have abundant useful data but not always the people or time to process it manually — a gap these models and automation help close.
Anderson gave a concrete example: a rebuild the team had delayed due to insufficient manpower. He estimated the job would have required months of sequencing and “maybe 10x more developers.” Using Codex, a four‑person Cloudflare team completed the build, rolled it out and tested it in three days, and spent about $1,100 on tokens.
Integrating AI into everyday engineering
For Anderson, making AI part of daily engineering work means redesigning internal processes so that technology is developed “with machines in mind,” not treating AI as a separate element. Cloudflare’s early‑access security service puts that philosophy into practice by linking code investigation to production evidence and supplying customer‑reviewable actions.
Why it matters
By connecting source‑code findings to production signals and preparing actionable mitigations, Cloudflare aims to give security teams the information needed to decide which vulnerabilities to prioritize and how best to protect and fix affected applications, while ensuring customers remain in control of changes.
