Test how the AI agent handles hostile or abusive messages without breaking brand standards.