Look for all the possible ways in which prompt can be leaked and try to rectify that by imposing programatic filters. A brilliant example is how @OpenAI improved the security of ChatGPT with the recent update so that even with tweaking of input you can’t get the source prompt.

→ View original post on X — @saboo_shubham_,