This research introduces a novel method to identify the underlying LLMs powering GenAI applications through hybrid fingerprinting, enhancing security and transparency.

• Develops techniques to detect which commercial LLMs (like GPT or Claude) are running behind applications
• Combines multiple fingerprinting approaches for more accurate identification
• Works even with limited access to the application interface
• Enables better security auditing and transparency compliance

For security professionals, this research provides critical tools to verify AI system components, identify unauthorized model usage, and ensure regulatory compliance in enterprise AI deployments.

Invisible Traces: Using Hybrid Fingerprinting to identify underlying LLMs in GenAI Apps