Problem
A Codex session in Air stops with the message "This request was blocked by our safety systems. Reason: Potentially unintended activity." The block can appear in the middle of a task, not only when you send a prompt.
Cause
The message comes from OpenAI, not from Air. OpenAI's misalignment monitoring reviews the agent's reasoning and actions in the background. It stops a conversation when it suspects that the agent misread the user's instructions, for example around sensitive data or destructive changes. Air shows the error that Codex returns.
Resolution
1) Start a new Codex session. OpenAI says a chat stopped by this check can't be resumed, and other users found that the same session kept blocking even simple follow-up questions.
2) If you think the block was a mistake, you can report it to OpenAI.
More about the check: Misalignment monitoring and Agent approvals and security.
Comments
0 comments
Please sign in to leave a comment.