[aw-failures] AI Moderator (Codex): chatgpt.com blocked by firewall

[github-actions]

Problem Statement

The AI Moderator workflow running with the Codex engine (v0.121.0) is attempting to make egress connections to chatgpt.com:443, which is blocked by the workflow's network firewall. This causes the agent job to fail silently — the agent starts (1.4 min runtime, 1 MB of stdio logged) but produces 0 parsed turns and the job concludes as failure.

This is a separate failure mode from the OpenAI 401 Unauthorized issue tracked in #27404.

Affected Workflow & Runs

• Workflow: AI Moderator
• Engine: Codex v0.121.0 (model: auto)
• Failed run: §24681803841
• Trigger: issues event (actor: verkyyi), 2026-04-20T17:28 UTC
• Parent investigation: [aw-failures] [aw] Failure Investigation Report — 2026-04-20 13:09–19:09 UTC #27410

Firewall Evidence

Domain	Allowed	Blocked
api.openai.com:443	13	0
chatgpt.com:443	0	1
github.com:443	2	0

OpenAI credentials were functional (13 successful API calls). The failure is a network-level block, not an auth issue.

Probable Root Cause

Codex v0.121.0 or a tool it invokes is making an outbound call to chatgpt.com during execution. Possible causes:

1. A browsing/search tool in the Codex agent's toolbox is calling chatgpt.com for a web lookup
2. The Codex binary itself contacts chatgpt.com for telemetry, licensing, or a capability check
3. A prompt or tool call in the workflow inadvertently triggers a chatgpt.com request

Proposed Remediation

1. Investigate: Read /tmp/gh-aw/aw-mcp/logs/run-24681803841/agent-stdio.log (1 MB) to find which tool or subprocess attempted the chatgpt.com call
2. If intentional: Add chatgpt.com:443 to the AI Moderator workflow's firewall allowlist
3. If unintentional: Fix the agent prompt or tool configuration to prevent the call, or upgrade/downgrade Codex to avoid the behavior

Success Criteria

• AI Moderator (Codex) completes without firewall blocks when triggered by the issues event
• Either chatgpt.com:443 is added to the allowlist and the call is confirmed intentional, or the call is eliminated from the agent's execution path

Generated by [aw] Failure Investigator (6h) · ● 285.2K · ◷

• expires on Apr 27, 2026, 7:18 PM UTC

[github-actions]

🍪 *Issue Monster selected this for Copilotecho BEGIN___COMMAND_OUTPUT_MARKER ; PS1= ; PS2= ; unset HISTFILE ; EC=0 ; echo ___BEGIN___COMMAND_DONE_MARKER___0 ; }

I've identified this issue as a good candidate for automated resolution and requested assignment to the Copilot coding agent.

If assignment succeeds, the Copilot coding agent will analyze the issue and create a pull request with the fix.

Om nom nom! 🍪

Caution

Security scanning requires review for Issue Monster

Details

Potential security threats were detected in the agent output. The workflow output should be reviewed before merging.

Review the workflow run logs for details.

🍪 Om nom nom by Issue Monster · ● 16.5K · ◷