One email check hides Claude’s Android debug menu

A teardown of Claude’s Android app published on GitHub on October 1 traced a hidden internal settings menu to a stored key, is_ant, which the app sets to true when the account’s email ends in @anthropic.com. The condition is local: in an APK the author rebuilt with it forced to pass, the entry appeared in Settings and the menu behind it opened intact, though the staging backend among its endpoint options still asked for a Cloudflare Access login.
Android release builds often go through R8 or a similar tool, which can remove unreachable code and shorten class, field and method names, making the app smaller and decompiled code harder to follow. Keys in SharedPreferences, Android’s store for small key-value settings, often stay readable through that step, and the author found about 100 of them in the app. In their Kotlin reconstruction, the code that writes is_ant runs during account setup and checks the email domain, which, the author writes, “strongly suggested is_ant was a local indicator for Anthropic employee accounts.” One of the places that reads the key decides whether the internal settings entry shows up in Settings.
On a rooted device the stored value could have been changed directly. Because they were exploring other parts of the app at the same time, the author repackaged it instead, first renaming the “New chat” label to confirm that the rebuilt, re-signed APK carried their edits.
The menu’s “Network Simulation” screen adds latency and forces request, upload and timeout failures. “Select API Endpoint” switches the app between Production, Staging, Localhost and a custom backend. “Show jank overlay” puts UI performance statistics on screen, connected to JankStats and Android’s FrameMetrics API, and the author says having those measurements inside the app makes a slow interaction easier to investigate while reproducing it. A dedicated screen tests push notifications, and an “Age Signal override” exposes controls for simulating different age signals. Turning on “Internal Settings bubble” adds a floating, draggable Ant button that opens the menu without a trip through Settings.
GrowthBook, which describes itself as an open-source feature flagging and experimentation platform, is integrated in the app as well, alongside an “Override feature flag” screen for changing local flag values and configurations. Those flags “can provide clues about experiments and possible future features,” the author notes, ranking the screen among the more interesting ones to explore.
The staging hostname was publicly reachable, but Cloudflare Access guarded it and demanded authentication. Cloudflare documents Access as deny by default: users who don’t explicitly match an Allow policy are denied.
“What impressed me most was the attention to detail, even in a part of the app intended only for internal use,” the author concludes.
Up next
OpenAI’s Dots run free until they open a Codex task
OpenAI’s Tibo Sottiaux says your primary Dot can work 24/7 without using any of your plan’s allowance. The meter starts once it hands a job to Codex, where that task draws usage as usual.
Related stories
- Claude desktop teardown: MCP servers run outside the VM
- Anthropic cuts deadline for critical bugs from 90 to 30 days
- HN post: Anthropic’s CVP Defense tier needs only KYC
- Asked to renumber pages, Claude deleted a syllabus
- A perfect refusal score isn’t a safer model, Anthropic says
- Testing Catalog: Claude iOS app gets “Saved logins” list
Comments
No comments yet. Be the first.
Join the conversation
Sign in to leave a comment. It appears after moderation.

