This seems like an issue similar to that which some organizations had with o1-preview: almost anything sent to the model by their organization being flagged and refused.
The input moderation is only supposed to inspect for attempts at extracting reasoning internals.
Self service: create a new project, new API key, and initially do not filter by any endpoint or model.
Don’t persist against these errors, as they can trigger automatic bans. You will also see there is no response_id to report.
Organization-specific issues will need be resolved by real account inspection by OpenAI staff, via “help” messages on the platform site that do not entertain a bot sending back inaction and advice.
