Know Thy Enemy: Securing LLMs Against Prompt Injection via Diverse Data Synthesis and Instruction-Level Chain-of-Thought Learning

Large language model (LLM)-integrated applications have become increasingly prevalent, yet face critical security vulnerabilities from prompt injection (PI) attacks. Defending against PI attacks faces two major issues: malicious instructions can be injected through diverse vectors, and injected instructions often lack clear semantic boundaries from the surrounding context, making them difficult to identify. To address these issues, we propose InstruCoT, a model enhancement method for PI defense that synthesizes diverse training data and employs instruction-level chain-of-thought fine-tuning, enabling LLMs to effectively identify and reject malicious instructions regardless of their source or position in the context. We evaluate InstruCoT across three critical dimensions: Behavior Deviation, Privacy Leakage, and Harmful Output. Experimental results across four LLMs demonstrate that InstruCoT significantly outperforms baselines in all dimensions while maintaining utility performance without degradation

Key Contributions

• Diverse training data synthesis covering multiple prompt injection vectors and injection positions to improve generalization of the defense
• Instruction-level chain-of-thought fine-tuning (InstruCoT) that teaches LLMs to explicitly reason about and reject malicious injected instructions
• Evaluation framework across three dimensions — Behavior Deviation, Privacy Leakage, and Harmful Output — benchmarked on four LLMs with no utility degradation

🛡️ Threat Analysis

Details

Similar Papers