LLM like humans can find vulnerabilities in black boxes. We already established 30 years ago that open source is usually more secure than closed source and that security by obscurity doesn't work.