GPU process crash permanently wedges desktop app (main process zombie); triggered by anti-bot fingerprinting page in Browser pane

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

Loading an SSRN page in the built-in Browser pane crashes Electron's GPU process
(exitCode 101457950). After the GPU crash the main process never recovers: it stops
logging entirely (60s [process-memory] ticks and hourly updater checks cease), never
relaunches the GPU process, all windows freeze, and the process lingers as a zombie
holding package file locks. Windows records no crash (no WER Event 1000/1001/1002)
because the process never exits — it deadlocks.

Reproduced 2/2. The second time happened immediately after a fresh OS reboot AND a
fresh reinstall of the app: session restore re-opened the Browser-pane tab and the
identical crash (same exitCode) followed 3 seconds later — i.e. session restore
creates a crash loop.

Secondary impact (why users end up "reinstalling"): the zombie main process blocks
MSIX servicing. AppXDeploymentServer logs 0x80073D02 ("apps must be closed", Id 638
naming Claude_pzs8sxrjxfjjc!Claude) 47 minutes after the freeze, and later installer
runs fail with 0x80070020 (sharing violation on app\resources\cowork-svc.exe) even
with ForceApplicationShutdownOption. The package was in fact healthy the whole time.

Control: four other Browser-pane previews the same day (one running 25 min) did not
crash — the trigger is page-specific (aggressive anti-bot / GPU-fingerprinting
scripts; GPU death follows ~1 s after the page's WebGPU requestAdapter call both times).

What Should Happen?

1. The main process should survive GPU-process death (relaunch the GPU process or

fall back to software rendering) and must never deadlock.

1. After a crash-triggered restart, the app should not blindly auto-restore

Browser-pane tab content that just killed the GPU process (crash-loop guard /
safe-restore mode).

Error Messages/Logs

main.log (local time, 2026-07-25):
19:22:13 [Preview] Created browser preview
19:22:17 GPU process gone: { type: 'GPU', reason: 'crashed', exitCode: 101457950, serviceName: 'GPU' }
-> zero further log lines until next app start at 20:18:33 (hourly updater check due 19:26:34 missing)
Identical second occurrence: 20:19:43 preview restore -> 20:19:46 same crash -> silence until 20:32:13

unknown-window.log (preview window console, ONLY at the two crash instants in the whole log):
OTS parsing error: Size of decompressed WOFF 2.0 is less than compressed size
%c%d font-size:0;color:transparent NaN   (repeated; devtools-detection trick)
WebGL: INVALID_ENUM: getInternalformatParameter ... EXT_color_buffer_float (~20x)
The powerPreference option is currently ignored when calling requestAdapter() on Windows. See https://crbug.com/369219127
A valid external Instance reference no longer exists. (Dawn/WebGPU)
WebGL: CONTEXT_LOST_WEBGL: loseContext: context lost

claude.ai-web.log: 19:22:17 CONTEXT_LOST_WEBGL (GPU death takes down all windows)

AppXDeploymentServer while wedged: 0x80073D02 (Id 638: running app Claude_pzs8sxrjxfjjc!Claude);
installer retries: 0x80070020 sharing violation on app\resources\cowork-svc.exe

Steps to Reproduce

1. In a desktop-app session, open the Browser pane and navigate to:

https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5207418
(SSRN sits behind aggressive anti-bot / GPU-fingerprinting scripts)

1. Within ~3-4 s the GPU process crashes and the whole app freezes permanently.
2. Restart the app -> session restore reloads the tab -> immediate identical re-crash.

Claude Model

Not sure / Multiple models

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

2.1.219 (bundled in desktop app 1.24012.9, MSIX)

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

Other

Additional Information

Environment: Windows 10 Pro for Workstations build 19045; MSIX package
Claude_1.24012.9.0_x64__pzs8sxrjxfjjc (Microsoft Store distribution); NVIDIA GPU
(driver nvlddmkm) — no driver-level TDR/reset events (System log Id 4101) during
either incident, so this is an in-process Electron GPU crash, not a driver reset.
Happy to provide full main.log / unknown-window.log excerpts on request.