Invisible prompts once tricked AI like old SEO hacks. Here’s how LLMs filter hidden commands and protect against manipulation ...