OpenAI Codex 0.144.5: improved dangerous-command detection with clearer rejection messages

What's new

• Expanded dangerous-command detection: added more rm forms — commands like rm -rf /, rm -rf ~ and other variants now more reliably rejected
• Clearer rejection messages: when Codex rejects a command, the message now includes a specific reason — not just generic operation rejected
• Security improvements follow Codex 0.144.4 (encrypted internal agent instructions)

Why it matters

Codex agents in automated pipelines can accidentally generate destructive shell commands — improved detection reduces the risk of unintended file-system operations. Clearer messages help with debugging when an agent hits limits.

How to try it

Codex 0.144.5 updates automatically — available for all paid Codex plans (Pro, Business, Enterprise).