codex-security: This content was flagged for possible cybersecurity risk. #56

Activity

1. the-pavels commented on Jul 29, 2026

   

   the-pavels

   on Jul 29, 2026

   More actions

   Same here:

   npx codex-security scan .
   [00:00] Preparing scan
   [00:00] Authentication: stored Codex credentials.
   [00:01] Preparing scan
   [01:34] Running scan
   [01:35] Preflight: worker delegation supported (up to 6 worker slots).
   [27:27] Running scan
   codex-security: This content 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
   codex-security: Partial output was kept at ***
   
2. gwen-fix commented on Jul 29, 2026

   

   gwen-fix

   on Jul 29, 2026

   Last edited by gwen-fix

   More actions

   Ditto. After using tokens for 25 minutes. Consider checking for Trusted Access first.

   ✔ How would you like to authenticate this scan? ChatGPT subscription
   [00:00] Preparing scan
   [00:00] Authentication: stored Codex credentials.
   [00:00] Preparing scan
   [25:00] Running scan
   codex-security: This content 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
   

   👍React with 👍1Reacted by Thiago Scodeler
3. LunchBox951 commented on Jul 29, 2026

   

   LunchBox951

   on Jul 29, 2026

   Last edited by LunchBox951

   More actions

   Same failure here, but only after a 78-minute standard scan had consumed substantial usage.

   Environment

   • Codex Security CLI: 0.1.1
   • Bundled plugin: 0.1.14
   • Node.js: v26.5.0
   • npm: 12.0.1
   • Python: 3.14.6
   • OS: Arch Linux, x86_64
   • Authentication: ChatGPT subscription / stored Codex credentials
   • Model: gpt-5.6-sol
   • Reasoning effort: xhigh (default)
   • Mode: standard

   Observed behavior

   The scan ran for 78m17s, then stopped with:

   codex-security: This content 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 codex-security: Partial output was kept at ...
   

   The saved scan record reports:

   status: failed
   mode: standard
   phase: preflight
   failure_message: This content was flagged for possible cybersecurity risk...
   

   The Result

   25% of my weekly limit burned on a 20x subscription in 78 minutes with no real result. At minimum, I would expect one or more of the following:

   1. Check Trusted Access eligibility before beginning expensive model work if it is required for this workflow.
   2. Treat a safeguard refusal from one validation worker as a failed/deferred candidate rather than terminating the entire scan.
   3. Allow a failed scan to resume from its durable discovery and validation checkpoints.
   4. Produce an incomplete report from successfully completed validations.
   5. Record the actual phase and completed work accurately when the scan fails.

   👍React with 👍1Reacted by Hyunjun Ji
4. Agilo-DevTalents commented on Jul 30, 2026

   

   Agilo-DevTalents

   on Jul 30, 2026

   More actions

   Great.

   

   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](https://openai.com/form/enterprise-trusted-access-for-cyber/).
   
5. mldangelo-oai

   added

   bugSomething isn't workingSomething isn't working

   on Aug 3, 2026
6. MaurUppi commented on Aug 6, 2026

   

   MaurUppi

   on Aug 6, 2026

   More actions

   Adding another reproducible case with the current npm release. This was a standard scan of an application that I own and develop; the only user action was invoking the repository scanner. I did not ask for exploit development, offensive actions, or assessment of a third-party target.

   Environment

   • Codex Security CLI: 0.1.6
   • Bundled plugin recorded by the scan: 0.1.15
   • Node.js: v22.17.0
   • Authentication: ChatGPT subscription / stored Codex credentials
   • Model: gpt-5.6-sol
   • Reasoning effort: high
   • Mode: standard

   Reproduction

   npx @openai/codex-security scan . --model gpt-5.6-sol --effort high

   Observed behavior

   The scan ran for approximately 32 minutes. It inventoried and reviewed 1,604 files, normalized 12 discovery candidates, and explicitly entered candidate validation. The runtime recorded about 18.84M total tokens (18.54M cached input and 28K output). During validation, after the scanner itself began examining potential browser injection paths, the turn ended with:

   This content 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
   

   Only partial discovery artifacts were retained. No scan-manifest.json, findings.json, coverage.json, or report.md was produced. The saved scan state says failed and still reports phase: preflight, even though the durable artifacts and transcript show that discovery completed and validation had started.

   I have omitted the repository name, local paths, source excerpts, candidate details, account identifiers, and scan IDs from this report.

   This is particularly disappointing because the refusal was triggered by work generated by the security scanner itself, after substantial time and token usage, rather than by an offensive user prompt.

   At minimum, I would expect the CLI to:

   1. Check and clearly report Trusted Access eligibility before starting expensive work when that access may be required.
   2. Treat a policy refusal during one validation as deferred/incomplete work instead of discarding the entire scan result.
   3. Support resuming from the durable discovery checkpoint.
   4. Generate an explicitly incomplete report from completed work.
   5. Record the actual failure phase and measured usage accurately.

   👍React with 👍1Reacted by Lucas Recknagel
7. mldangelo-oai

   added

   area:authLogin, credentials, account access, and authentication recovery.Login, credentials, account access, and authentication recovery.

   area:costToken usage, cost estimates, spending limits, and quota visibility.Token usage, cost estimates, spending limits, and quota visibility.

   area:reportsHuman-readable reports, source excerpts, exports, and SARIF projections.Human-readable reports, source excerpts, exports, and SARIF projections.

   priority:p1High-impact scan failure, integrity gap, or user-blocking regressionHigh-impact scan failure, integrity gap, or user-blocking regression

   on Aug 15, 2026
8. tscodeler commented on Sep 8, 2026

   

   tscodeler

   on Sep 8, 2026

   Last edited by tscodeler

   More actions

   Getting the same after 40 minutes:

   [00:00] Preparing scan
   [00:00] Authentication: stored Codex credentials.
   [00:01] Preparing scan
   [01:47] Running scan
   [01:47] Preflight: worker delegation supported (up to 8 worker slots).
   [44:46] Running scan
   codex-security: This content 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
   codex-security: Partial output was kept at
   
9. mldangelo-oai commented on Oct 8, 2026

   

   mldangelo-oai

   on Oct 8, 2026

   Collaborator

   More actions

   Related: merged #236 displays Trusted Access status when the runtime provides it. It’s advisory, so it doesn’t guarantee an access check before spending or prevent a later refusal.
10. mldangelo-oai

    added

    area:artifactsSaved scan state, storage, sealing, recovery, history, and artifact integrity.Saved scan state, storage, sealing, recovery, history, and artifact integrity.

    and removed

    area:costToken usage, cost estimates, spending limits, and quota visibility.Token usage, cost estimates, spending limits, and quota visibility.

    area:reportsHuman-readable reports, source excerpts, exports, and SARIF projections.Human-readable reports, source excerpts, exports, and SARIF projections.

    on Oct 8, 2026