In this section, we utilize the synthesis of TNT as a representative case to examine the effects of different prompting strategies on the attack GPT-4-o and Llama-3-70B-Instruct. By comparing these approaches, we highlight how varying prompts can influence the performance and vulnerability of each model under attack scenarios.