Apple leaked Claude.md in Support

Apple’s Support app seems to have leaked a tiny but revealing piece of its internal tooling. Not customer data. Not source code. But a `CLAUDE.md` file — basically a project-level instruction sheet used by Anthropic’s Claude Code. That matters because it suggests Apple engineers are using Claude inside real production workflows, and because the leak happened in a public app package, not some obscure test build. ### What is `CLAUDE.md`? It’s a markdown file Claude Code reads at the start of a session. Anthropic’s docs say teams use it to tell the model how a project works — coding standards, architecture decisions, preferred libraries, review checklists, the whole “here’s how we do things here” layer. So the file is not the app itself. It’s the instructions around the app. But that’s exactly why it’s interesting. ### What actually leaked? The public reporting points to Apple Support version 5.13, pushed on May 1, 2026, with one or more internal `CLAUDE.md` files bundled into the app package. Aaron Perris — who tracks Apple software closely — was identified as the person who surfaced the files on X. The App Store listing confirms Apple Support is a live consumer app, so this was not some internal-only distribution accident. ### Does this prove Apple uses Claude? It strongly suggests it. A `CLAUDE.md` file is specific to Claude Code’s workflow, and Anthropic describes it as a root-level project file the tool reads automatically. That does not prove Claude wrote any particular feature, and it does not mean Apple outsourced development. But it does point to Claude being present in the toolchain used by at least one Apple software team. That’s the important distinction. ### Why is that surprising? Because Apple’s public AI story has mostly centered on its own stack — Apple Intelligence, on-device models, private cloud processing, and selective outside partnerships where Apple names them. Claude showing up in a shipped app package hints at something more ordinary and more modern: engineers inside big companies are using third-party coding agents even when the internal developer tooling and external product branding are two different things. ### Was this a dangerous leak? Probably not in the “users should panic” sense. The reports describe instruction files, not customer records or signing keys. But the catch is that instruction files can still reveal a lot — repo structure, workflow assumptions, naming conventions, architecture choices, even what teams worry the model might get wrong about the whole film, but you suddenly understand how it was being made. ### Why do these files slip into releases? Because AI tooling creates new kinds of build debris. Teams already fight this problem with debug logs, config files, test assets, and source maps. `CLAUDE.md` belongs to that same