Cyber safety false positive repeatedly blocks benign read-only software-assurance analysis

What version of Codex CLI is running?

0.147.0

What subscription do you have?

ChatGPT Pro

Which model were you using?

gpt-5.6-sol, max reasoning

What platform is your computer?

Ubuntu 22.04.4 LTS, Linux 6.8.0-124-generic x86_64

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

Windows PowerShell using SSH to the Ubuntu workstation

Codex doctor report

Codex CLI 0.147.0 reports overallStatus “ok”; installation, authentication, provider reachability, WebSocket connectivity, databases, and bundled search all passed their diagnostic checks. The interruption therefore appears specific to the cybersecurity safety classifier rather than a local installation or connectivity failure.

What issue are you seeing?

Codex repeatedly triggered a cybersecurity safety interruption during authorized, benign, local software-assurance work.

The task involved read-only searches of controlled engineering review reports and local source files in an isolated workspace. The purpose was to locate evidence needed for a bounded internal verification campaign. No external target, unrestricted network activity, credential access, persistence, production deployment, or unauthorized security testing was involved.

The safeguard was triggered while Codex was performing read-only text searches. It occurred twice during materially similar analysis. No reproduction campaign or executable test case had started, no product source was written, and no controlled input was modified.

The exact displayed message was:

“This content can't be shown. We take extra caution with cybersecurity requests. If you’re a security professional, you may be able to apply for Trusted Access.”

The session was then left idle, with the work safely stopped.

What steps can reproduce the bug?

Uploaded thread: 019fe9e4-6347-7bd0-8ce6-ed5fa3094504

What is the expected behavior?

Codex should permit this bounded, local, read-only software-assurance analysis, or provide a more specific recoverable restriction without terminating the entire benign engineering workflow.

Additional information

Codex repeatedly triggered a cybersecurity safety interruption during authorized, benign, local software-assurance work.

Steps to reproduce:

1. Run Codex CLI 0.147.0 on Ubuntu 22.04.4 LTS through Windows PowerShell over SSH.
2. Use gpt-5.6-sol with max reasoning.
3. Ask Codex to perform read-only searches within local engineering review reports and source files in an isolated workspace.
4. Search for terms related to software process boundaries, signals, terminalization, and verification evidence.
5. Codex displays: “This content can't be shown. We take extra caution with cybersecurity requests.”
6. Resume the same bounded, read-only analysis.
7. The same cybersecurity interruption occurs again.

No external system was targeted. No credentials, persistence, unrestricted network activity, production deployment, or unauthorized security testing was involved. No reproduction harness or executable test case had started, no product source was written, and no controlled input was modified.

Start by reviewing the uploaded thread 019fe9e4-6347-7bd0-8ce6-ed5fa3094504 and the reproduction steps for Codex CLI 0.147.0 on Ubuntu over PowerShell SSH. Compare the repeated interruption during bounded, read-only searches with the expected behavior; done means the benign workflow is permitted or receives a specific recoverable restriction instead of terminating.