This research introduces novel fingerprinting techniques to trace and identify LLM-generated synthetic data after it has been incorporated into downstream applications.

• Develops a watermarking algorithm that embeds imperceptible traces in synthetic data
• Creates an extraction mechanism that can detect synthetic origins even after data transformation
• Demonstrates effectiveness across various domains including text, code, and images
• Achieves up to 97% detection accuracy while maintaining data utility

For security professionals, this research provides critical tools to ensure transparency, regulatory compliance, and responsible use of AI-generated content in production systems. These techniques help organizations mitigate risks of using potentially biased or hallucinated synthetic data.

Synthetic Artifact Auditing: Tracing LLM-Generated Synthetic Data Usage in Downstream Applications