[BUG] GPU process crashes app when Browser pane blocks a subresource ("PreviewContext" / private-resolving host)

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?

Opening a URL in the Browser pane (preview_start) and interacting with the page reliably
crashes the desktop app's GPU process, taking down the whole app instantly. Reproduced
twice in one session with identical log signatures:

[warn] [PreviewContext] Blocked subresource to private-resolving host { resourceType: 'xhr' }
[info] GPU process gone: { type: 'GPU', reason: 'crashed', exitCode: 101457950, serviceName: 'GPU' }

Both crashes: same exit code (101457950), same trigger sequence. Immediately after the GPU
process dies, the app closes entirely and won't reopen without running a repair
(Settings > Apps > Claude > Advanced options > Repair on Windows).

Steps to reproduce:

1. Have Claude call preview_start with any URL to open the Browser pane.
2. Interact with the loaded page (any follow-up page tool call).
3. The Browser pane's PreviewContext blocks a subresource request to a private-resolving

host, the GPU process crashes, and the whole app goes down.

Log location: %LOCALAPPDATA%\Claude\Logs\main.log

Environment:

• OS: Windows 11 Pro (10.0.26200)
• App: Claude desktop, MSIX install, appVersion 1.40609.0
• Package: Claude_pzs8sxrjxfjjc

Expected: The Browser pane should not crash the GPU process / whole app when a subresource
request gets blocked by the private-network security check — that should just fail the one
request, not take down the renderer.

What Should Happen?

The desktop app should remain stable when Claude uses Browser pane tools
(preview_start, page interaction, console reading, etc.) instead of crashing.

Error Messages/Logs

From %LOCALAPPDATA%\Claude\Logs\main.log — identical signature on two separate crashes:

[warn] [PreviewContext] Blocked subresource to private-resolving host { resourceType: 'xhr' }
[info] GPU process gone: {
  type: 'GPU',
  reason: 'crashed',
  exitCode: 101457950,
  serviceName: 'GPU'
}

Occurrence 1: 2026-09-01 20:26:01 (app didn't relaunch until 20:46:26, ~20 min later)
Occurrence 2: 2026-09-01 21:35:29 (app didn't relaunch until 21:37:15, ~2 min later)

Both times the exit code is identical (101457950), and both are immediately preceded by
the same "Blocked subresource to private-resolving host" warning from PreviewContext
(the Browser pane's internal component name).

Steps to Reproduce

1. In an active Claude Code session with the Browser pane available, have Claude call

preview_start with any URL (e.g. a live production website) to open a page in the
Browser pane.

1. Have Claude follow up with any page-interaction tool call (e.g. reading console

messages, taking a screenshot/wait action, or navigating).

1. The Browser pane's PreviewContext blocks a subresource request to a private-resolving

host (a normal Private Network Access security check) — but instead of just failing
that one request, it takes the GPU process down with it.

1. The GPU process crash brings down the whole desktop app instantly. It does not

reopen on its own; running "Repair" via Settings > Apps > Claude > Advanced options
is required before it will launch again.

Reproduced twice in one session, both with the identical log signature above.

Claude Model

None

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

2.1.247 - The desktop app shell itself is on 1.40609.0 if the form wants that instead.

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

Terminal.app (macOS)

Additional Information

_No response_