This research introduces a novel framework that simultaneously optimizes prompt security and system performance in edge-cloud LLM systems, addressing the rising threat of prompt engineering-based attacks.

• Employs a two-phase detection mechanism to identify malicious prompts before they cause harm
• Creates a security-aware scheduler that balances robust security with minimal latency
• Achieves 99.4% attack detection rate while maintaining high system efficiency
• Demonstrates practical implementation across diverse edge-cloud configurations

Why it matters: As LLMs become increasingly deployed at the edge, robust security measures are essential to prevent privacy leaks and resource wastage without compromising performance—filling a critical gap in current enterprise AI security strategies.

Joint Optimization of Prompt Security and System Performance in Edge-Cloud LLM Systems