[Chrome/Computer Use] campus.163.com is falsely blocked while campus.game.163.com works

What version of the Codex / ChatGPT app are you using?

• ChatGPT/Codex desktop app: 26.818.31338
• Codex CLI: 0.149.0-alpha.4
• Bundled Chrome Browser Use plugin: 26.818.31338
• Computer Use runtime: 26.819.1000816

What platform is your computer?

macOS 26.5.2 (Build 25F84), Apple Silicon (arm64).

What issue are you seeing?

Browser Use and Computer Use appear to falsely block the normal NetEase campus recruiting site campus.163.com, even though the related public recruiting site campus.game.163.com works normally.

The public job page works:

https://campus.game.163.com/app/job/position?id=102

Opening Personal Center redirects to:

https://campus.163.com/app/personal/apply?tab=net

The page renders normally when opened manually in Chrome. The Chrome extension is connected, the site approval setting is Always allow, and Browser Use can enumerate the existing tab and read its title and URL.

However, claiming the tab and requesting a DOM snapshot fails immediately:

Browser Use rejected this action due to browser security policy.
Reason: The site-safety policy blocks this action; no user permission prompt or Auto-review was attempted.
Browser use is not permitted on https://campus.163.com/app/personal/apply.

Computer Use is also stopped after Chrome reaches the same URL:

This session has been stopped because Computer Use is not allowed on the current browser URL.
Note that Computer Use is not allowed on this URL even if the user navigates to it themselves.

This is a read-only request to view the user's own campus application status. No form submission, upload, purchase, messaging, or account modification was requested.

Steps to reproduce

1. Enable the official Chrome Browser Use extension in Codex.
2. Set website approval to Always allow.
3. Open:
   https://campus.game.163.com/app/job/position?id=102
4. Observe that Browser Use / Computer Use can inspect and interact with this page.
5. Open Personal Center, which redirects to:
   https://campus.163.com/app/personal/apply?tab=net
6. Call browser.user.openTabs(); the tab title and URL are returned successfully.
7. Claim the tab and request a DOM snapshot.
8. Observe site_status_blocked before any DOM content is returned.
9. Alternatively, use Computer Use after the redirect and observe that the session is stopped because of the current browser URL.

Expected behavior

Codex should allow user-authorized, read-only access to this legitimate recruiting site, or present a recoverable permission prompt.

If the URL is intentionally restricted, the UI should provide an actionable explanation. A site-status decision for one path should also not unintentionally block unrelated paths on the same hostname.

Actual behavior

The URL is centrally blocked before DOM access. The user's Always allow setting does not override the result, and no permission prompt or Auto-review is attempted.

Additional diagnostics

Local diagnostics indicate that Browser Use checks the current URL with OpenAI's Aura site-status service before executing the DOM command.

The request includes the full path and query string, but the Browser Use client caches the result by normalized hostname for 24 hours. This may cause a path-specific decision to overblock other pages on campus.163.com.

The failed tool result contains only the URL-policy error. It contains no target-site HTTP status, response headers, DOM, body, or screenshot, which indicates that the observed failure happens before the requested page inspection.

Computer Use's local runtime also contains an AuraSiteStatusURLPolicyChecker, consistent with both browser surfaces applying the same class of URL-policy check.

Related general report: #29343

Feedback ID: 01a0327f-6ae5-7791-9621-62819048115f

Related task: 01a029cd-bdfa-7542-b05d-0c40d8905881

Start with the AuraSiteStatusURLPolicyChecker and the browser.user.openTabs flow described in the report. Trace how the full URL is checked and how normalized-hostname caching affects campus.163.com paths. Done means the legitimate read-only page is allowed or gets a recoverable prompt without unintentionally blocking unrelated paths.