We propose a new class of fine-tuned LLMs, Permissioned LLMs, that enforce access control on responses to queries, thus protecting sensitive training/tuning data from unauthorized queries.