The paper introduces INSTRUCTIONALFINGERPRINT® (IF), a novel method for fingerprinting large language models (LLMs) to protect intellectual property and ensure compliance with licensing terms. IF leverages lightweight instruction tuning to implant a confidential private key as a backdoor, causing the LLM to generate specific text when the key is present. The method is designed to be harmless, lightweight, and persistent even after extensive downstream fine-tuning. The study evaluates IF on 11 popular LLMs, demonstrating its effectiveness, robustness, and efficiency. IF supports multi-stage fingerprinting, similar to the MIT License, and prevents publisher overclaim. The approach is applicable to both white-box and black-box scenarios, making it versatile for various use cases. The paper also discusses the limitations and ethical considerations of the method.The paper introduces INSTRUCTIONALFINGERPRINT® (IF), a novel method for fingerprinting large language models (LLMs) to protect intellectual property and ensure compliance with licensing terms. IF leverages lightweight instruction tuning to implant a confidential private key as a backdoor, causing the LLM to generate specific text when the key is present. The method is designed to be harmless, lightweight, and persistent even after extensive downstream fine-tuning. The study evaluates IF on 11 popular LLMs, demonstrating its effectiveness, robustness, and efficiency. IF supports multi-stage fingerprinting, similar to the MIT License, and prevents publisher overclaim. The approach is applicable to both white-box and black-box scenarios, making it versatile for various use cases. The paper also discusses the limitations and ethical considerations of the method.