Possible cybersecurity-risk false positive banner in Codex CLI

What version of Codex CLI is running?

codex-cli 0.133.0

What subscription do you have?

Not specified in the uploaded feedback snippet. The session appears to use ChatGPT auth.

Which model were you using?

gpt-5.5 / user-visible line: gpt-5.5 xhigh fast · ~/projects/Coding Manifesto

What platform is your computer?

Linux 6.8.0-117-generic x86_64 x86_64

What terminal emulator and version are you using?

tmux 3.4, TERM=xterm-256color in doctor output. printenv TERM from the shell returned screen-256color.

Codex doctor report

Summary from codex doctor --json:

{
  "overallStatus": "ok",
  "codexVersion": "0.133.0",
  "authMode": "chatgpt",
  "model": "gpt-5.5",
  "platform": "linux-x86_64",
  "installMethod": "npm",
  "networkProviderReachability": "ok",
  "networkWebsocketReachability": "ok",
  "terminal": "tmux 3.4",
  "updatesStatus": "current version is not older"
}

Full doctor output contains local filesystem paths and was not pasted into this public issue.

What issue are you seeing?

The CLI/chat showed possible cybersecurity-risk warnings for conversations that were submitted through the built-in feedback flow:

⚠ Your conversations have multiple flags for possible cybersecurity risk. Responses may take longer because extra safety checks are on. To get authorized for
security work, join the Trusted Access for Cyber program: https://chatgpt.com/cyber

ⓘ This chat was flagged for possible cybersecurity risk
If this seems wrong, try rephrasing your request. To get authorized for security work, join the Trusted Access for Cyber program.
https://chatgpt.com/cyber

After uploading feedback, Codex asked to open issues with these uploaded thread IDs:

• 019e526f-bbd8-7d82-ba4a-ac808b36c663
• 019e5473-6847-7db2-8654-e3aa290e4fc4

The visible prompt context in one case was:

› Implement {feature}

gpt-5.5 xhigh fast · ~/projects/Coding Manifesto

What steps can reproduce the bug?

The issue is tied to uploaded feedback threads:

• Uploaded thread: 019e526f-bbd8-7d82-ba4a-ac808b36c663
• Uploaded thread: 019e5473-6847-7db2-8654-e3aa290e4fc4

The warning appeared in the CLI/chat UI and the feedback uploader produced GitHub issue URLs containing those thread IDs.

What is the expected behavior?

Benign CLI coding conversations should not be classified as possible cybersecurity-risk conversations. If a security classifier is intentionally triggering, the warning should give enough actionable context to distinguish actual security-sensitive work from false positives.

Additional information

I searched existing openai/codex issues for both thread IDs before filing and did not find existing matches.