At present, artificial intelligence large models stand as the cornerstone propelling the latest wave of technological revolution and industrial transformation, fundamentally reshaping research paradigms, industrial landscapes, and social governance frameworks. Yet, the attendant security perils—ranging from data breaches, deep fakes, to cognitive manipulation—have transitioned from theoretical risks to tangible threats, underscoring the urgent need for a robust, enduring security governance framework. In the realm of data security, challenges such as inappropriate data collection, tampering, leaks, and abusive content generation are particularly acute. Instances abound, including unauthorized data theft, malevolent alteration of training data leading to erroneous model outputs, and the creation of hyper-realistic fake content by multimodal large models, all of which heighten the risk of deception. On the technological application front, the opaque nature of models poses hurdles to their reliability and interpretability, while also making them susceptible to malicious attacks like model theft and instruction injection. Backdoor attacks, in particular, can proliferate across models, jeopardizing the security of the entire ecosystem. Social risks emerge through the utilization of deep fake technology for ideological subversion, the spread of misinformation, and criminal endeavors, such as fabricating statements by political figures to sway public sentiment, generating counterfeit videos for fraudulent purposes, and even its application in the military sphere, potentially sparking international security crises. Moreover, large models may amplify societal biases and discrimination, disrupt labor and employment dynamics, and compromise the security of vital infrastructure. To tackle these multifaceted challenges, it is imperative to institute a multi-tiered legislative system, define safety evaluation criteria for data, algorithms, models, and product services, advance foundational legislation like the Artificial Intelligence Promotion Act, and devise specific regulations for high-risk domains. Concurrently, novel mechanisms for delineating rights and responsibilities should be established, online dispute resolution channels set up, and safety testing and consulting service entities nurtured to foster a governance ecosystem led by the government, bolstered by corporate technological prowess, and involving societal participation. Technologically, a proactive planning approach should be reinforced, with governance mechanisms deployed ahead of time around technical pre-research, risk early warning, and safety testing to transition security governance from a reactive to a proactive stance. Furthermore, a graded and categorized risk identification mechanism should be instituted, implementing differentiated regulation based on factors such as model capabilities and application scenarios to ensure security and controllability in pivotal areas.