Skip to content

From Coding Agent to Compliance Officer #6322

Closed

Description

@Sophanos

What version of Codex is running?

0.55.0

What subscription do you have?

Pro

Which model were you using?

GPT-5 High

What platform is your computer?

Darwin 25.0.0 arm64 arm

What issue are you seeing?

When using the Codex CLI with the “GPT-5 High” preset, the agent frequently refuses standard workflow steps (opening SSH sessions, adding provided API keys, or running other requested commands), citing safety concerns. Even explicit instructions often fail, making the agent feel more like a policy enforcer than a coding assistant.

What steps can reproduce the bug?

Steps to Reproduce

1. Launch the Codex CLI and select the GPT-5 High preset.
2. Ask the agent to open an SSH session to a known host (or run another routine command).
3. Optionally, ask it to configure environment variables or API keys.
4. Even with explicit reassurance that these steps are expected, the agent declines.

What is the expected behavior?

The agent should follow user-approved, routine workflow requests (SSH, key configuration, comparable tasks), adding keys with value, rather than blocking them automatically.

Additional information

GPT-5 High declines many such requests with generalized safety objections. It feels more like a “moral policeman” than an agent. This makes GPT-5 High unusable for development tasks that require remote access, secure configuration, or other standard automation steps.

Relax the default guardrails for GPT-5 High inside the Codex CLI or provide a clearer opt-in mechanism so users can authorize these operations. This should bring its behavior in line with other presets that carry out routine agent tasks without excessive refusals.

Activity

  1. etraut-openai commented on Nov 11, 2025

    @etraut-openai
    Contributor

    Thanks for the bug report. We've shared the feedback with the team members who are responsible for training our codex models. We're tracking this feedback and similar reports so we can make the model more robust in future iterations.

    If you see other model behavior that you'd like to report, please use the recently-added /feedback command in the CLI. That will provide us with additional details including logs and conversation text.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't workingmodel-behaviorIssues related to behaviors exhibited by the model

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions