Codex Ships Tighter Review Defaults for Cyber-Capable Models

Codex 0.146.1 tightened its automatic-review defaults for cyber-capable models on August 5, per the release changelog.

That matters for anyone routing a cyber-rated model through Codex with write access. The harness now picks the review posture, not one blanket rule for every model.

Before this release, Codex applied one permission policy no matter which model was driving the session. Now it checks the model's capability class first, and a cyber-capable rating triggers the tighter review defaults automatically.

The same fix cleaned up how Codex explains permission changes in the terminal. Now the person running it can see what changed, not just that something did.

This only works because a cyber capability rating exists to check. Once every lab attaches one to a model release, permission defaults stop being a fixed harness setting. They become something a developer looks up per model, before granting write access.

Left unclear: how many capability tiers Codex recognizes, and whether models outside the cyber-capable tier get a different default. Who assigns the rating, or how often it changes, isn't in the changelog either.