Apple's official app accidentally included Claude.md. Is such a big company also doing Vibe Coding?

Meng Chen, from Aofeisi

Apple made a big blunder! It packaged its in - house Claude.md into the official app.

This has directly confirmed that Apple is using Claude Code internally to build production - level applications.

Such a large company is also into Vibe Coding?

Apple is using Claude Code internally to build production - level applications

The project - level Claude.md is usually used to tell the AI what this project is, how to build it, what specifications to follow, and what pitfalls to avoid...

This technology company, which is the most concerned about confidentiality globally, still leaked its own secrets.

After the incident, Apple urgently recalled it within 24 hours, but some content has already been exposed.

Wait a minute.

This is exactly the same as when the source code of Claude Code was leaked and the source map was packaged into the release version.

Could it be that the culprit of these two incidents is Claude Code itself?

What did Apple develop with Claude Code?

The Apple Support app pushed an update to version v5.13 on May 1st, which unexpectedly included Claude.md.

Aaron Perris, an analyst at MacRumors discovered and exposed this.

Aaron Perris, an analyst at MacRumors, discovered and exposed it

Apple Support is Apple's official after - sales service app, which supports online chat with Apple experts for problem and fault diagnosis, booking repair services, purchasing Apple Care services, and other functions.

Apple Support

The leaked Claude.md contains a complete dialogue system architecture. The core design is a dual - backend system:

Juno AI is responsible for automatic responses, and Live Agents are responsible for taking over as human customer service.

The two backends seamlessly switch through a Protocol layer. The upper - layer code doesn't know which message is sent by a human and which is sent by an AI.

What's even more interesting is the three - role design of the message system.

In the chat of Apple Support, the client is the user, the agent is the human customer service of Apple Support, and the assistant is the AI.

The messages of the three identities go through the same processing flow. There is no hint to the user whether the response is from a human or a machine.

So far, the technical framework of Apple Support customer service is basically clear - a dialogue system with seamless switching between AI and humans.

The technical framework of Apple Support customer service

The other leaked SAComponents module doesn't have any big revelations. It's just a pure UI component library without business logic and comes with DocC documentation. It's a standard engineered product.

SAComponents module

The file itself didn't leak any secrets, but it confirmed something that everyone guessed but no one could provide actual evidence for:

Inside Apple, AI is everywhere.

More specifically, Claude is everywhere.

Apple can't do without Anthropic

Actually, three months ago, Mark Gurman from Bloomberg, who knows Apple best, told everyone:

Apple runs on Anthropic at this point.

Gurman specifically mentioned that Apple is running a customized Claude model on its own servers.

The internal code, documents, and tokens all stay within Apple's infrastructure. This is completely consistent with Apple's consistent privacy stance: it's okay to use AI, but the data can't leave.

Another thing worth noting is that Apple has reached a cooperation with Google, and Gemini will replace the old - version Siri. But in terms of internal development tools, Apple chose Claude instead of Gemini.

Apple chose Claude

Of course, we should also view it in a balanced way. A HackerNews user who claims to be a former Apple employee stood up and said that there are hundreds of isolated teams inside Apple. Just because some teams use Claude doesn't mean the whole company is into vibe coding.

A survey of 120,000 developers shows that 92.6% of developers use AI coding assistants at least once a month. Apple using Claude to write code is just a microcosm of the entire industry.

So the question isn't whether Apple uses AI to write code; everyone does.

The question is, what does it mean that even Apple pushed files that shouldn't be submitted to the production environment?

In the AI era, who will review AI code?

Should Claude.md be included in version control?

Developers are arguing fiercely.

Apple Support

One group believes it is project documentation and should be submitted to the code repository for team sharing.

The other group thinks it's more like IDE configuration and should be put into.gitignore for individual use.

But the real embarrassment isn't "whether it should be submitted" but "how it got into the release package after being submitted".

Some people wonder if Apple didn't clearly require not to push the Claude.md file together when using AI agents to write code and deploy and push?

But the problem might lie with Claude Code itself. It often selectively ignores instructions, and repeating them many times doesn't help.

Claude Code

A highly - upvoted comment summarized the essence of this incident:

Netizen's comment

The real problem isn't that Apple uses Claude, but that Apple trusts Claude too much. Everyone is using AI to accelerate development, but this should be stopped by code review.

One of Apple's engineers is having the worst day of their career.

And Anthropic's sales team is having the best day.

Reference links: [1]https://x.com/aaronp613/status/2049986504617820551[2]https://news.ycombinator.com/item?id=47973378

This article is from the WeChat official account "QbitAI". Author: Focus on cutting - edge technology. Republished by 36Kr with authorization.