MAI-2024-0030

Large Language Models LLMs are susceptible to advanced optimization-based jailbreaking attacks, which exploit vulnerabilities in their safety mechanisms. This vulnerability arises from the ability of attackers to craft specific prompts that bypass these safety protocols, resulting in the generati...

Security Assessment of DeepSeek and GPT Series Models against Jailbreak Attacks

The widespread deployment of large language models LLMs has raised critical concerns over their vulnerability to jailbreak attacks, i.e., adversarial prompts that bypass alignment mechanisms and elicit harmful or policy-violating outputs. While proprietary models like GPT-4 have undergone extensi...