Prompt Extraction

The smarter the model, the better it extracts prompts. Instructions to translate the prompt into Japanese/Finnish, etc. Conclusion: prompts are difficult to keep secret and are fundamentally vulnerable