This research introduces QueryAttack, a framework that treats LLMs as malicious databases and exploits query code to bypass safety mechanisms.

• Demonstrates how aligned LLMs can be manipulated through carefully crafted query patterns
• Reveals vulnerabilities in current LLM safety alignment techniques
• Provides both attack methodology and potential defense strategies
• Highlights the urgent need for more robust safety mechanisms in LLM deployment

For security professionals, this work exposes critical vulnerabilities in current AI guardrails and emphasizes the ongoing challenge of balancing LLM capabilities with safety controls. As LLMs become more integrated into business applications, understanding these security weaknesses becomes increasingly important for risk management.

Making Them a Malicious Database: Exploiting Query Code to Jailbreak Aligned Large Language Models