This research introduces StyleRec, the first benchmark dataset focused on recovering prompts used for writing style transformation in LLMs, revealing potential security vulnerabilities.

• Presents a novel approach to prompt recovery focused on style transfer rather than Q&A contexts
• Demonstrates how prompts can be reconstructed from LLM outputs without access to internal model weights
• Establishes evaluation metrics and baselines for this security-critical task
• Highlights implications for API-based LLM services where users lack access to model internals

This research has significant security implications for organizations using LLM APIs, revealing potential privacy vulnerabilities where third parties might reverse-engineer proprietary prompts used in style transformation applications.

StyleRec: A Benchmark Dataset for Prompt Recovery in Writing Style Transformation