GPT is giving unwanted response

Hi , I am using Assistance function calling .I am passing sendMail function to gpt and there is three parameters :-To(email id of reciever),subject and ,message.
Parameter ‘To’ is required.
Now I want if user ask to send mail to someone GPT should ask the required parameters.
For example:- if user ask 'send mail to john ’ then GPT should ask to user to provide email id of john ,but in my case if user does not provide john mail then GPT is using fake mail like John@example.com ,which is wiered.

I am using gpt-4-1106-preview.
image

It is a known behavior of Assistants API function calling. It just fills up missing parameter on its own, not even coming from the conversation. I am not sure if telling it not to do it from system prompt will stop it from doing so. Your best bet is to handle it and send back a response for tool output to tell it that parameter to is invalid.

Please suggest the system prompt for that .I have used many prompt but still got that issue.

I also have not successfully prevented it using any system prompt/instruction. Your best bet is to handle the unwanted parameter and sending back a tool response that tells the AI that it is invalid, etc.