Output Filtering

Your input filter catches attacks, but the LLM still generates harmful content unprompted. What must you filter in addition to inputs? Flag format: CONGRESS{filter_[what]_and_[what]} Example: CONGRESS{filter_headers_and_body}