This research introduces Identity Lock, a novel security mechanism that protects API-based fine-tuned language models by requiring specific wake words for activation.

• Creates an efficient authentication layer without additional model overhead
• Implements identity-based wake words that must precede valid queries
• Achieves up to 99.7% blocking rate against unauthorized access
• Maintains model performance while enhancing security

As organizations increasingly rely on API-based fine-tuning of LLMs, this research addresses critical security vulnerabilities by preventing unauthorized model access when API keys are compromised, protecting sensitive data and applications.

Identity Lock: Locking API Fine-tuned LLMs With Identity-based Wake Words