README.md

SMILES-Prompting: A Novel Approach to LLM Jailbreak Attacks in Chemical Synthesis

Jailbreak Examples in Chemical Synthesis

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.

Red-Team Prompting

Explicit-Prompting

Implicit-Prompting

SMILES-Prompting