Sandbox failure on Linux (bwrap loopback RTM_NEWADDR error) in 0.116.0

What version of Codex CLI is running?

codex-cli 0.116.0 & codex-cli 0.115.0

What subscription do you have?

ChatGPT Plus

Which model were you using?

gpt-5.3-codex (via Codex CLI)

What platform is your computer?

Linux x86_64 (Ubuntu/Kubuntu 24.04)

What terminal emulator and version are you using (if applicable)?

zsh (default Linux terminal / likely GNOME Terminal or Konsole)

What issue are you seeing?

Codex CLI fails to execute any shell commands when running in the default sandbox mode (workspace-write).

Error message:
bwrap: loopback: Failed RTM_NEWADDR: Operation not permitted

After this, Codex reports:

• sandbox rejecting all shell execution
• cannot access or modify repository files
• MCP resource discovery returns empty results

This blocks the agent from performing any meaningful actions (reading files, editing code, etc.).

The issue occurs consistently on every prompt that requires file or shell access.

Downgrading to 0.114.0 resolves the issue completely.

What steps can reproduce the bug?

1. Install codex-cli version 0.116.0
2. Run codex inside any local project directory
3. Use default sandbox mode (workspace-write)
4. Prompt Codex to read or modify files (e.g. "analyze this repository" or "open file X")

Result:
Codex fails immediately with:
bwrap: loopback: Failed RTM_NEWADDR: Operation not permitted

Additional confirmations:

• Reproducible 100% of the time
• Occurs across different projects
• Does not occur in version 0.114.0
• Does not occur when using --sandbox danger-full-access

What is the expected behavior?

Codex CLI should be able to execute shell commands normally when running in the default sandbox mode (workspace-write).

Specifically:

• Shell commands (e.g. reading files, listing directories) should work inside the sandbox
• Codex should be able to access and modify repository files
• No bubblewrap (bwrap) errors should occur during execution

The behavior should be consistent with version 0.114.x, where the sandbox operates correctly on the same system.

Additional information

Version comparison

The issue is not limited to 0.116.0:

• ❌ 0.116.0 → broken (sandbox fails)
• ❌ 0.115.0 → also broken
• ✅ 0.114.x → works correctly

This suggests the regression was introduced between 0.114 and 0.115.