Skip to content

False possitve cyber policy rejection #53688

Open

Description

@codealxcubed

What issue are you seeing?

I’m seeing an apparent false-positive cyber_policy rejection during a routine rewrite of my application's backend. The work consists of implementation, refactoring, and tests.

The provider returns HTTP 400 with:

This content was flagged for possible cybersecurity risk.

Details:

  • Model: gpt-6.1-sol
  • Access: ChatGPT subscription Responses service
  • Occurrences: October 13, 2026, at 08:43:51 UTC and 14:38:00 UTC
  • The session completed 14 successful provider rounds between the two rejections.
  • I’ve encountered the same class of rejection in Codex as well.

The rejection interrupts a long-running development goal. The error does not identify what triggered the classification or provide enough information to distinguish a false positive from an actual concern. The saved goal remains intact, but work stops.

What steps can reproduce the bug?

This happens on a / goal when I'm rewriting my backend to rust.

What is the expected behavior?

The model does it's job and not get blocked by some arbitrary safety check.

Additional information

No response

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't workingsafety-checkIssues related to safety and abuse checks

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions