Cloudflare 运营着全球最大的网络之一，代理了约 20% 的网络流量。各组织依赖该网络来确保其应用和服务的安全、快速和可靠。
Cloudflare 将通过 Daybreak 访问的 OpenAI 模型与其全球网络和威胁上下文相结合，应用于 漏洞发现与修复（在新窗口中打开），这是一项通过 Cloudflare Managed Defense 提供的早期访问服务。这些模型会调查代码并评估潜在漏洞。Cloudflare 加入实时流量数据、安全信号和验证，以对发现结果进行优先级排序，并为客户准备拟议的缓解措施和补丁以供审查。
Codex 将 OpenAI 的角色扩展到 Cloudflare 的整个工程组织，开发人员使用它来提高产出并改善代码质量。
“Daybreak 和 Codex 提供的自动化之美在于，你可以分配机器去完成繁琐的工作，这样人类就可以对最重要的工程和安全决策行使判断力。”
—Mark Anderson，Cloudflare 收入总裁
Cloudflare 客户故事视频。
使用 Daybreak 扩展安全分析
安全团队面临着源源不断的漏洞信息和警报。手动筛选这些数据需要时间和精力，导致团队用于调查和解决关键问题的能力减少。“传统安全团队拥有大量有用的数据。他们只是没有足够的人力和时间来处理这些数据，”Anderson 说。
Cloudflare 的方法是将源代码中的漏洞与生产环境中正在发生的事情联系起来。受影响的路径是否活跃？它是否正在接收可疑流量？现有安全控制是否已经在保护它？这些信号为安全团队提供了更多上下文，以便决定首先调查什么。
通过 漏洞发现与修复（在新窗口中打开），Cloudflare 使用通过 Daybreak 访问的 OpenAI 模型来调查客户授权的代码、识别潜在漏洞，并检查支持每项发现的证据。该服务准备拟议的代码补丁，并在适当情况下准备自定义 Web 应用防火墙规则，以在开发人员审查修复方案期间减少暴露。Cloudflare 对这些提议运行自动化检查，其团队在与客户分享之前会进行审查。
对于 Cloudflare 客户而言，好处是从识别漏洞到保护受影响应用的清晰路径。该服务为他们提供证据以对漏洞进行优先级排序、拟议的边缘保护以减少暴露，以及可供审查的代码修复，同时他们保留对实施哪些更改的控制权。
使用 Codex 在 3 天内重建产品
保护应用还取决于团队编写和维护的代码。Anderson 将 Cloudflare 对 Codex 的使用与更高的开发人员生产力、更好的代码质量以及开发过程中引入的更少错误联系起来。
“在 Cloudflare，我们公司三分之一的人是软件开发者，为我们的平台构建服务和功能，他们 100% 都在使用 Codex 来提高产出并交付更好的代码。”
—Mark Anderson，Cloudflare 收入总裁
Anderson 描述了团队最近推迟的一项重建工作。“我们之所以推迟构建它，是因为我们没有足够的人力来完成，”他说。随着 Codex 扩展开发人员生产力，问题从“我们能承担这个任务吗？”转变为“为什么不呢？”
Anderson 估计，这项工作原本需要数月的排序和“可能多 10 倍的开发人员”。借助 Codex，Cloudflare 一个由四名开发人员组成的团队承担了这项构建工作。“我们在三天内构建完成，推出并测试，花费了 1,100 美元的 token 费用，”他说。
围绕 AI 和人类判断塑造网络防御
对于 Anderson 来说，让 AI 成为日常工程的一部分意味着“重建内部流程，以机器的思维来构建技术，不是将其作为单独的元素，而是作为集成到构建过程中的东西。”
Cloudflare 的早期访问安全服务通过将代码调查与生产证据以及客户可以审查的拟议行动联系起来，将这种方法付诸实践。目标是让安全团队获得所需信息，以决定优先处理哪些漏洞以及如何保护和修复受影响的应用，同时客户保留对更改的控制权。
Cloudflare operates one of the world’s largest networks and proxies roughly 20% of the web’s traffic. Organizations depend on that network to keep their applications and services secure, fast, and reliable.
Cloudflare combines OpenAI models accessed through Daybreak with its global network and threat context in Vulnerability Discovery and Remediation(opens in a new window), an early-access service offered through Cloudflare Managed Defense. The models investigate code and assess potential vulnerabilities. Cloudflare adds live traffic data, security signals, and validation to prioritize findings and prepare proposed mitigations and patches for customers to review.
Codex extends OpenAI’s role across Cloudflare’s engineering organization, where developers use it to increase output and improve code quality.
“The beauty of the automation that Daybreak and Codex provide is that you can assign machines to do the grunt work so humans can exercise their judgment on the most important engineering and security decisions.”
—Mark Anderson, President of Revenue, Cloudflare
Cloudflare customer story video.
Scaling security analysis with Daybreak
Security teams face a constant stream of vulnerability information and alerts. Manually sifting through that data takes time and effort, leaving teams with less capacity to investigate and address critical issues. “Traditional security teams have a ton of useful data. They just don’t have the people and time to deal with it,” Anderson says.
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? Those signals give security teams more context for deciding what to investigate first.
Through Vulnerability Discovery and Remediation(opens in a new window), Cloudflare uses OpenAI models accessed through Daybreak to investigate customer-authorized code, identify potential vulnerabilities, and check the evidence supporting each finding. The service prepares proposed code patches and, where appropriate, custom web application firewall rules to reduce exposure while developers review a fix. Cloudflare runs automated checks on these proposals, and its team reviews them before sharing them with customers.
For Cloudflare customers, the benefit is a clear path from identifying a vulnerability to protecting the affected application. The service gives them evidence to prioritize vulnerabilities, proposed edge protections to reduce exposure, and code fixes to review, while they retain control over which changes to implement.
Rebuilding a product in 3 days with Codex
Protecting applications also depends on the code teams write and maintain. Anderson connects Cloudflare’s use of Codex to greater developer productivity, better code quality, and fewer bugs introduced during development.
“At Cloudflare, a third of our company are software developers building services and features for our platform, and 100% of them use Codex to boost their output and deliver better code.”
—Mark Anderson, President of Revenue, Cloudflare
Anderson describes a recent rebuild the team had postponed. “We were putting off building it because we didn’t have the manpower to do it,” he says. With Codex scaling developer productivity, the question shifts from “Can we take this on?” to “Why not?”
Anderson estimates that the work would have required months of sequencing and “maybe 10x more developers.” With Codex, a Cloudflare team of four developers took on the build. “We built it in three days, rolled it out and tested it, and spent $1,100 on tokens,” he says.
Shaping cyber defense around AI and human judgment
For Anderson, making AI part of everyday engineering means “rebuilding the internal process to build technology with machines in mind, not as a separate element, but as something that’s integrated into the build process.”
Cloudflare’s early-access security service puts that approach into practice by connecting code investigation with production evidence and proposed actions customers can review. The goal is to give security teams the information they need to decide which vulnerabilities to prioritize and how to protect and fix affected applications, while customers retain control over changes.
本文内容采集自官方网站，排版和翻译可能与原页面存在差异。
阅读官方全文