Apple shipped Claude.md configuration files inside the Apple Support app update v5.13, accidentally revealing that they use Anthropic's Claude Code agent to build production software. The files, spotted by developer Aaron on X, are the setup documents that tell Claude Code how to navigate a codebase, follow coding standards, and produce acceptable output. Think of it as an onboarding manual for an AI teammate. Apple rushed out v5.13.1 to strip the files, but the damage was done.
Bloomberg's Mark Gurman has reported that Apple runs custom versions of Claude on its own servers, so the discovery isn't shocking if you've been paying attention. Still, seeing the actual configuration files land on millions of iPhones is something else. The files contained directives around Swift, SwiftUI, Human Interface Guidelines compliance, accessibility standards, and performance optimization for iOS. This is a structured, repeatable workflow where Claude Code functions as a defined member of the engineering team.
The irony wasn't lost on anyone. Apple recently took a stance against "vibe-coded apps" on the App Store, and here they are shipping an app partially built with exactly that class of tool. One commenter on X put it simply: "You may not make vibe-coded apps, but Apple can support you with vibe-coded updates." Questions about quality control are fair. If an AI agent can push code that makes it into a shipped update without a human catching a stray configuration file, what else slips through?
Apple is one of the most secretive companies in tech. Claude Code isn't a pilot program for them. It's infrastructure, baked into how at least one team ships software. An AI agent pushed code into an Apple update that landed on millions of iPhones, and nobody caught the stray config file until a developer scrolling through X spotted it. If that can happen at Apple, with their legendary QA, it's happening everywhere.