Claude Desktop (Windows/MSIX): GPU process crashes app on Cloudflare Turnstile-protected site (marinha.mil.br)
Summary
Claude Desktop crashes reproducibly (the Electron/Chromium GPU process dies and the whole app restarts) when using the "Browser pane" tool to navigate to a site protected by Cloudflare Turnstile (anti-bot challenge). After repeated crashes, Windows starts requiring "Repair" via Settings > Apps > Advanced options before the app can be reopened. The problem persists even after repairing or fully reinstalling the app.
Environment
- App: Claude Desktop, version 1.28929.0 (isPackaged: true)
- Install method: MSIX (
Claude_1.28929.0.0_x64__pzs8sxrjxfjjc) - Node: 24.18.0
- OS: Windows 11 Home Single Language, build 10.0.26200, x64
- Free memory at crash time: ~1.3-3 GB out of 16 GB (system under memory pressure)
Steps to reproduce
- In Claude Code Desktop, open the Browser pane tool (
preview_start) pointed at a URL. - Navigate to
https://www.marinha.mil.br/chn-6/(redirects tohttps://marinha.mil.br). - The site serves a Cloudflare Turnstile challenge page ("Um momento... Executando verificação de segurança"), which loads
challenges.cloudflare.com/turnstile/v0/...and spawns multiple Web Workers fromblob:URLs. - Within seconds, the app's GPU process crashes and the whole app restarts on its own.
Expected behavior
The browser pane should handle this kind of page (or fail in an isolated/graceful way, without taking down the GPU process / whole app).
Actual behavior
Reproducible GPU process crash, followed by an automatic full app restart. Happened at least twice in the same session, with the identical exit code:
Excerpt 1 - logs/main.log, 2026-08-12 09:21:11:
GPU process gone: {
type: 'GPU',
reason: 'crashed',
exitCode: 101457950,
serviceName: 'GPU'
}
(automatic app restart logged right after, 09:23:27)
Excerpt 2 - logs/main.log, 2026-08-12 10:26:26:
GPU process gone: {
type: 'GPU',
reason: 'crashed',
exitCode: 101457950,
serviceName: 'GPU'
}
(automatic app restart logged right after, 10:27:23)
Console errors observed on the page right before the crash (via the console-reading tool):
[error] Failed to load resource: the server responded with a status of 403 ()
[info] Creating a worker from 'blob:https://www.marinha.mil.br/...' violates the following Content Security Policy directive: "script-src 'unsafe-inline' 'unsafe-eval'" (report-only, non-blocking)
Additional impact
After repeated crashes, manually reopening Claude Desktop started showing:
"You'll need to access advanced options for Claude and select Repair. If you're still having issues with it, reinstall this application or contact your administrator."
The problem persisted even after repairing the app via Windows' advanced app options and after a full uninstall/reinstall.
Note
The crash appears specific to the GPU process under load from the anti-bot challenge (proof-of-work / fingerprinting in Web Workers), not a content/policy block - there is no refusal message or policy error anywhere in the app, only the GPU crash.
This issue has 2 comments on GitHub. Read the full discussion on GitHub ↗