Grok Bot computer egress (Cloudflare, via local proxy 127.0.0.1:8791) intermittently geolocates to China (Shanghai / Zhengzhou). US-region APIs (e.g. TypeSafe / Jev) then return HTTP 451 (“not available in your region”). When the same egress lands in Los Angeles, the APIs work. Please pin Grok Bot computer egress to the US.
Environment
Product: Grok Bot computer (shared box)
Egress path: Cursor-managed Cloudflare exit (AS13335), traffic via local proxy 127.0.0.1:8791
“Route traffic through this computer” is OFF on all of my devices (confirmed after the earlier Chrome ERR_TIMED_OUT fix)
Affected API: TypeSafe / Jev — HTTP 451 when egress is China; succeeds when egress is US (Los Angeles)
Timeline (Asia/Shanghai)
2026-10-11 ~09:50 — Repeated HTTP 451; egress IPs in 104.28.208.x / 104.28.240.x geolocate to Shanghai (also Zhengzhou earlier that day)
2026-10-11 ~10:32 — Brief recovery; egress Los Angeles
Stable US egress for the Grok Bot computer so US-only APIs keep working.
Actual
Egress flips between China (Shanghai/Zhengzhou, 104.28.208.x / 104.28.213.x / 104.28.240.x) and US (Los Angeles, 104.28.195.190). China egress → HTTP 451 from TypeSafe/Jev.
Ask
Please pin Grok Bot computer Cloudflare egress to the United States (or otherwise keep it out of China geolocation). I cannot change this from the client; “Route traffic through this computer” is already off everywhere.
Open Grok Bot and leave “Route traffic through this computer” OFF on all devices.
From the Bot computer, call a US-region API (TypeSafe / Jev) via the local relay on 127.0.0.1:8765 (egress still goes through Cursor’s Cloudflare proxy at 127.0.0.1:8791).
Check the Bot computer’s public IP geolocation (e.g. ipinfo).
When egress geolocates to China (Shanghai / Zhengzhou, e.g. 104.28.213.205 / 104.28.245.202), the API returns HTTP 451 “not available in your region”.
When the same path later geolocates to Los Angeles (e.g. 104.28.195.190), the identical API call succeeds with no code change.
Repeat over several days: egress flips between China and US without any client setting change.
There’s no setting to pin the Grok Bot computer to a specific country right now, and we can’t change that for an individual account. Its location isn’t fixed, so TypeSafe and Jev return 451 whenever it shows up somewhere they don’t serve. “Route traffic through this computer” doesn’t affect this, so it’s fine to leave it off.
We’ve let the team know about your request for a region option, and I’ll post here if there’s an update.
Same issue here, with evidence that narrows down the cause. Team BAD Grok Leads (id 36811780), ticket T-I30654.
Two seats, identical settings (Route traffic OFF on every device, same registered desktop, same admin policy) - The [email protected] computer exits via Fastly, Washington DC (2a04:4e41::/32), loc=US.
The [email protected] computer (box boot id d8b9741c-c806-43ae-962c-e69563a1f1d4) exits via Cloudflare 2a09:bac1::/32 and 2a09:bac5::/45, labelled Lebanon (cdn-cgi trace: colo=IAD, loc=LB, warp=off). It was on the Fastly path too until the platform recreated the computer overnight Oct 23-24.
What we measured on the [email protected] computer (shell and browser, SAND_HTTP_PROXY_NAME=runner-egress-proxy, local proxy port 8791) - The label follows our registered desktop’s apparent country with a 30-60 minute lag. With a US VPN on the desktop, the pool’s addresses become US-labelled (ipwho.is and the Cloudflare trace both say US); with the VPN off, they drift back to Lebanon within a couple of hours. Traffic never transits the desktop: with the app fully closed, the box still exits through the same pool.
Toggle cycles, full app restarts, sign-out/sign-in, curl -4/-6 and the explicit proxy all exit the same pool.
So on this computer the Cloudflare pool is geolabelled after the member’s device, while the [email protected] computer gets Fastly US. For unattended US-only research this makes Google render in Arabic and Cloudflare-protected US sites block the bot.
Request: move the [email protected] computer onto the Fastly/US path like [email protected], or tell us how to get a computer that is not on the device-tracking pool. Support (T-I30654) says no action is possible; this thread says a region option was logged on Oct 14. Happy to share full probe outputs.
Update with new data (ticket T-I30654). Two seats on the same team, same PC, “Route traffic through this computer” OFF on both, identical in-box egress config (SAND_HTTP_PROXY_NAME=runner-egress-proxy, SAND_EGRESS_TUNNEL_ENABLED=1, proxy 127.0.0.1:8791). Seat A runs on SAND_BOX_CLUSTER=us11 (snapshot …block_v1) and exits Cloudflare WARP (2a09:bac1/bac5); seat B runs on us10 (snapshot …v5) and exits Fastly 2a04:4e41 Washington, always US. Seat A’s geolabel follows wherever the Grok desktop app last connected from: with the app running behind a US VPN it turned US within about an hour and stayed US for 12+ hours after the app was closed, while web visits to grok.com (even through a US proxy) never changed it. The app’s main process connects straight to AWS us-east-1 and ignores --proxy-server. Support says it can’t move the computer, so: could staff move seat A to the us10/Fastly path, or stop the WARP label from following the desktop app?