That's the result of guard rails on the hosted service. They run checks on the query before it even hits the LLM as well as ongoing checks at the LLM generates

That's the result of guard rails on the hosted service. They run checks on the query before it even hits the LLM as well as ongoing checks at the LLM generates output. If at any moment it detects something in its rules, it immediately stops generation and inserts a canned response. A model alone won't do this.