I started to build an iOS client for the "Open"AI API and, while playing with the end-points and the "stream" option, I managed to send an empty prompt. Imagin

I started to build an iOS client for the "Open"AI API and, while playing with the end-points and the "stream" option, I managed to send an empty prompt. Imagine my surprise when I got back instructions on how to install Roblox. While I understand that any data sent doesn't fall under GDPR protection and the prompts on the chat interface might be used for future training, I was under the impression, based on their data usage policy (https://openai.com/policies/api-data-usage-policies https://openai.com/policies/api-data-usage-policies) that they don't use data submitted via the API for training. After this incident, I'm having second doubts. On a somewhat related topic, would it be possible to send a prompt to the API in an obfuscated manner, which would be unreadable by a human, but that the neural network can parse (and reply to)?