This research introduces a novel jailbreaking framework that successfully bypasses LLM safety measures to generate malicious content through segmented prompts.

• Divides harmful prompts into innocuous segments processed in parallel
• Achieves up to 92% success rate in bypassing safety filters
• Tests 500 malicious prompts across 10 cybersecurity categories
• Demonstrates critical security vulnerabilities in current LLM defenses

Security Implications: This work exposes significant vulnerabilities in existing LLM safety mechanisms, showing how malicious actors could generate harmful code while evading detection. The findings emphasize the urgent need for more robust, attack-resistant safety measures in AI systems.

Prompt, Divide, and Conquer: Bypassing Large Language Model Safety Filters via Segmented and Distributed Prompt Processing