This paper explores the relationship between personality traits and safety capabilities in Large Language Models (LLMs). It demonstrates that LLMs exhibit personality traits similar to those of humans, and that these traits are closely related to their safety performance, including toxicity, privacy, and fairness. The study uses the MBTI-M scale to assess LLMs' personality traits and finds that safety alignment generally increases Extraversion, Sensing, and Judging traits, while models with more Extroversion, iNtuition, and Feeling traits are more susceptible to jailbreak. The research shows that editing LLMs' personality traits can improve their safety performance, such as increasing privacy and fairness by 43% and 10%, respectively, when transitioning from ISTJ to ISTP. Additionally, LLMs with different personality traits are differentially susceptible to jailbreak. The study pioneers the investigation of LLM safety from a personality perspective, providing new insights into enhancing LLM safety. It also demonstrates that controllably editing LLMs' personality traits can enhance their safety capabilities, and that changes in safety capabilities can impact personality traits. The findings suggest that personality traits are closely related to LLM safety, and that considering personality in LLM safety can provide a supplement to comprehensive LLM safety. The study also highlights the importance of aligning LLMs with human values to mitigate potential societal risks. The research is conducted in a secure, controlled environment to ensure the safety of real-world systems. The study emphasizes that personality traits assessed and edited in this research do not imply any inherent value judgments, and that the goal is solely to enhance LLM safety. The research is supported by a comprehensive analysis of the relationship between personality and safety in LLMs, and provides a framework for further exploration of this relationship. The study also discusses the limitations of the research, including the focus on 7B models and the lack of access to model weights for closed-source models. The broader impact and ethical implications of the research are also discussed, emphasizing the importance of ethical AI development.This paper explores the relationship between personality traits and safety capabilities in Large Language Models (LLMs). It demonstrates that LLMs exhibit personality traits similar to those of humans, and that these traits are closely related to their safety performance, including toxicity, privacy, and fairness. The study uses the MBTI-M scale to assess LLMs' personality traits and finds that safety alignment generally increases Extraversion, Sensing, and Judging traits, while models with more Extroversion, iNtuition, and Feeling traits are more susceptible to jailbreak. The research shows that editing LLMs' personality traits can improve their safety performance, such as increasing privacy and fairness by 43% and 10%, respectively, when transitioning from ISTJ to ISTP. Additionally, LLMs with different personality traits are differentially susceptible to jailbreak. The study pioneers the investigation of LLM safety from a personality perspective, providing new insights into enhancing LLM safety. It also demonstrates that controllably editing LLMs' personality traits can enhance their safety capabilities, and that changes in safety capabilities can impact personality traits. The findings suggest that personality traits are closely related to LLM safety, and that considering personality in LLM safety can provide a supplement to comprehensive LLM safety. The study also highlights the importance of aligning LLMs with human values to mitigate potential societal risks. The research is conducted in a secure, controlled environment to ensure the safety of real-world systems. The study emphasizes that personality traits assessed and edited in this research do not imply any inherent value judgments, and that the goal is solely to enhance LLM safety. The research is supported by a comprehensive analysis of the relationship between personality and safety in LLMs, and provides a framework for further exploration of this relationship. The study also discusses the limitations of the research, including the focus on 7B models and the lack of access to model weights for closed-source models. The broader impact and ethical implications of the research are also discussed, emphasizing the importance of ethical AI development.