Linked Presentation: Instruction Backdoor Attacks Against Customized LLMsQuantifying Privacy Risks of Prompts in Visual Prompt Learning