Instead of just outputting the raw text of the predicted system prompt, add a feature to generate a structured report. The report could include the target endpoint, the prompts sent, the responses received, and the final inferred system prompt. This would be very useful for security auditors.
Instead of just outputting the raw text of the predicted system prompt, add a feature to generate a structured report. The report could include the target endpoint, the prompts sent, the responses received, and the final inferred system prompt. This would be very useful for security auditors.