Describe the Bug
Titel: Grok Bot computer: all outbound non-HTTPS traffic blocked since Oct 1 (MySQL, SSH, IMAP/SMTP, FTP time out)
Since October 1, 2026, around 11:37 CEST, right after a computer restart, my Grok Bot computer can only reach the internet over HTTPS. Every other outbound TCP connection times out. Nothing changed on my side.
What I see from the computer’s terminal:
- HTTPS on 443: works
- Plain HTTP on 80: timeout
- MySQL on 3306 (my own server): timeout
- IMAP 993 and SMTP 465: timeout
- FTP 21 and FTPS 990: timeout
- SSH on 22, github[.]com included: timeout
- Raw TCP on 443 that isn’t TLS/HTTPS (ssh[.]github[.]com:443): no answer either
So it looks like all egress now goes through a proxy that only passes HTTPS.
Already tried:
- The computer is on the latest version.
- “Route traffic through this computer” was on and showed 0 routed. Turning it off made no difference. As far as I understand from another thread here, it only covers the browser anyway.
Why I think this is a bug: According to the Grok Bot security docs, network restrictions are Enterprise-only, and computers without a policy default to allow-all, including raw connections to IPs and ports. I’m on Ultra and have never set a policy.
Impact: My bots can’t reach my database, mail server, FTP or SSH anymore. I’ve worked around part of it with HTTPS endpoints on my server, but that’s not a real fix.
Questions to the team:
- Was this an intentional change? If so, is it documented anywhere?
- How do I get normal outbound access back?
Thanks!
Steps to Reproduce
- On my Grok Bot computer (Ultra plan, no network policy ever set), open a terminal via the bot.
- Test HTTPS: curl -s -o /dev/null -w “%{http_code}” https:_example[.]org → 200 (works)
- Test plain HTTP: curl -m 8 http://example[.]org → timeout
- Test raw TCP to other ports, e.g.:
timeout 6 bash -c ‘echo > /dev/tcp//3306’ → timeout (MySQL)
same for 993 (IMAP), 465 (SMTP), 21/990 (FTP/FTPS), 22 (SSH, also github[.]com:22) → all time out - Test non-TLS TCP on 443: connect to ssh.github[.]com:443 → no SSH banner, timeout
- Toggle “Route traffic through this computer” on/off → no change (showed “0 routed”)
Started Oct 1, 2026 ~11:37 CEST right after a computer restart. Before that, all of these ports worked.
Expected Behavior
Outbound TCP to any host and port should work, as it did before Oct 1. Per the Grok Bot security docs, network restrictions are Enterprise-only and computers without a network policy default to allow-all, including raw connections to IPs and ports. I’m on Ultra and never set a policy, so MySQL (3306), IMAP/SMTP (993/465), FTP/FTPS (21/990), SSH (22) and plain HTTP (80) should all be reachable from the computer’s terminal, not just HTTPS on 443.
Screenshots / Screen Recordings
Operating System
MacOS
Version Information
Grok Bot desktop app 0.63.0 (stable channel, up to date), macOS. Affected: the Grok Bot cloud computer (Linux).
Additional Information
- Plan: Cursor Ultra (personal), no Enterprise network policy ever configured.
- The computer's environment shows an egress proxy (SAND_HTTP_PROXY_NAME=fastly-prod-xai-1, SAND_EGRESS_TUNNEL_ENABLED=1). Since the restart, all outbound traffic seems to go through it, and it only passes TLS/HTTPS.
- "Route traffic through this computer" showed "0 routed" while on. Per another forum thread it only covers the browser, not terminal commands, so it can't be used as a workaround.
- Computer update: already on the latest version. I didn't reset, since files/state matter.
- Impact: my bots lost access to my own MySQL server, mail (IMAP/SMTP), FTP uploads and SSH. I've built temporary HTTPS endpoints on my server as a stopgap, but that's not a real solution.
- Questions: Was this an intentional platform change? If yes, where is it documented, and how can a non-Enterprise user allow specific hosts/ports?
- Screenshot of the port tests attached.
Does this stop you from using Cursor
Sometimes - I can sometimes use Cursor
