Patients are increasingly turning to LLMs and AI systems for medical guidance before traditional healthcare channels, but these systems do not simply generate answers; they actively shape recommendations by ranking, filtering, and prioritizing medical information, providers, and treatments. This introduces a hidden recommendation layer inside AI systems where outcomes are influenced by training data, retrieval pipelines, system prompts, and safety alignment layers. In healthcare, this creates a critical security and governance challenge, as these recommendation pathways are not transparent or auditable. The talk explores how LLM recommendation stacks work, where control points exist, and why emerging risks such as prompt injection, content poisoning, and recommendation manipulation directly impact patient safety. It also highlights the governance gap in AI-driven healthcare decisions, raising the core question of who ultimately controls what AI recommends and how that control is exercised.