This research introduces Identity Lock, a novel security mechanism that protects fine-tuned LLMs against unauthorized API access without requiring expensive re-training.

• Creates a secure wake word system that validates user identity before allowing model access
• Prevents exploitation even when API keys are compromised
• Demonstrates effectiveness across multiple domains including healthcare
• Offers a practical solution for resource-limited organizations using API-based fine-tuning

For security professionals, this approach provides a critical layer of protection that goes beyond passive watermarking by actively preventing unauthorized use of valuable fine-tuned models.

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