之江实验室ZhejiangLab基础理论研究院人工智能与安全团队生成式大模型安全与隐私白皮书作者:徐晓刚，吴慧雯，刘竹森，李想，涂文轩，梁伟轩，张毅，刘哲版权归之江实验室所有欢迎交流2023年6月6日ThedevelopmentoftheGenerativeAI,e.g.,LargeLanguageModels(LLM),havebeenpopularinbothacademicandindustrialcommunitiesonaworldwidescale,especiallytheChatGPTseries.ThesuccessofChatGPTandGPT4hasshownthefuturedirectionofdevelopingAGI.However,largegenerativemodelsalsosufferfromtheissueofdata/modelsecurityandprivacy.Weshouldnotethatlargegenerativemodelswouldbringalotofsecurityandprivacyproblems,whentheydemonstrategreatpowerinchangingourlife,suchasdataleakingandthepropagationoffakenews...