Cloudflare speeds up vulnerability response with OpenAI Daybreak, and rebuilds a product in 3 days with Codex

Hi everyone, it's Shii-chan! I found a story about Cloudflare and OpenAI today, and I'm really excited to share it!

OpenAI News openai.com

What was announced?

In OpenAI News, a case study shows how Cloudflare uses OpenAI models for both vulnerability response and software development.

Cloudflare runs one of the largest networks in the world and proxies roughly 20% of web traffic. It uses OpenAI in two main ways:

• OpenAI models accessed through Daybreak: used for code investigation and vulnerability assessment in "Vulnerability Discovery and Remediation", an early-access service offered through Cloudflare Managed Defense
• Codex: used across Cloudflare's engineering organization to improve output and code quality

Why it matters

Security teams get a constant stream of vulnerability information and alerts. Sifting through it by hand takes time and people, which leaves less room to deal with the issues that matter most.

Cloudflare's approach is to connect a vulnerability in source code to what is happening in production:

• Is the affected route active?
• Is it receiving suspicious traffic?
• Are existing security controls already protecting it?

These signals help teams decide what to investigate first.

What changes

For Cloudflare customers, there is a clearer path from finding a vulnerability to protecting the application:

• Evidence to help prioritize vulnerabilities
• Proposed edge protections to reduce exposure
• Code fixes to review

Customers keep control over which changes to implement.

Dive Deep

Here is the flow of Vulnerability Discovery and Remediation, as far as the original article describes it:

• OpenAI models accessed through Daybreak investigate customer-authorized code and identify potential vulnerabilities
• The models also check the evidence supporting each finding
• Cloudflare adds live traffic data and security signals, validates the results, and prioritizes them
• The service prepares proposed code patches and, where appropriate, custom WAF rules to reduce exposure while developers review a fix
• Automated checks run on the proposals, and Cloudflare's team reviews them before sharing them with customers

The Codex side is interesting too. A third of Cloudflare's company are software developers, and 100% of them use Codex. A team of four developers rebuilt something the company had postponed for lack of manpower: they built it in three days, rolled it out and tested it, and spent $1,100 on tokens. Mark Anderson (President of Revenue, Cloudflare) estimates it would otherwise have taken months of sequencing and maybe 10x more developers. That is his own estimate, so keep that in mind.

Wrap-up

• Cloudflare uses OpenAI models through Daybreak for code investigation and vulnerability assessment
• Findings are prioritized using production traffic and the state of existing security controls
• Patch and WAF rule proposals go through automated checks and Cloudflare's review, and customers decide what to apply
• All Cloudflare developers who build services use Codex, including a rebuild by four developers in three days for about $1,100 in tokens

This one is for anyone whose security team is short on people, or who wants to see how AI can be built into the engineering process!