This research introduces a robust testing framework to determine if one language model is derived from another, addressing critical intellectual property and security concerns in AI.

• Establishes reliable methods to verify model provenance in production environments
• Protects intellectual property by detecting unauthorized model derivatives
• Enables identification of affected models when vulnerabilities are discovered in foundation models
• Creates accountability in the AI supply chain through technical verification

For security professionals, this framework offers crucial tools to manage risk as models proliferate through fine-tuning and adaptation. By establishing clear provenance, organizations can better enforce licensing terms, track vulnerabilities, and ensure compliance across their AI ecosystems.

Model Provenance Testing for Large Language Models