Why are simple prompts flagged as violating policy?

The symptom from over a year ago was a hyperactive and broken moderation system when reasoning models were introduced, where particular organizations couldn’t send anything without a policy warning, and then OpenAI also banning the accounts and stealing the credits of innocents. You have something intermittent, and it is likely about the particular content in combination with the developer prompt being used. It’s possible that if there was a flagged category for “code”, it would be for tasks evaluated as cyber threats, such as vulnerability scans being done on a code base.