New capabilities for building safe models include watermarking, prompt refining, and prompt debugging and work with any large language models. If you use Python for accessing API endpoints or web ...