In the 2024 government work report, the “AI+” initiative was proposed for the first time, clearly accelerating the formation of a new quality of productivity driven by artificial intelligence. Over the past decade, China’s AI technology has developed rapidly, successfully applied in manufacturing, energy, transportation, finance, education, and healthcare, laying a solid foundation for the smooth implementation of the “AI+” initiative. The implementation of the “AI+” initiative not only reflects China’s high regard for new generation information technologies such as AI but also represents a profound layout for future economic and social development. AI will play an increasingly important role in promoting China’s industrial upgrading and facilitating the development of new quality productivity.AI technology has tremendous potential in advancing nuclear science, technology development, and application. China places great importance on and supports the research and development of intelligent nuclear energy technologies. The Ministry of Science and Technology’s major project on “Next Generation Artificial Intelligence” supports the construction of demonstration application scenarios, building a full-chain, whole-process AI industry application ecosystem. The application of these demonstration scenarios not only enhances the level of intelligence across various industries but also provides technological support and demonstration effects for the development and application of AI technologies in the nuclear energy sector. The “14th Five-Year Plan for Modern Energy System” explicitly states the promotion of AI applications in nuclear power and the active construction of intelligent energy systems. Currently, there is a clear path to further promote the development and application of AI in the nuclear field, implementing the “AI+” initiative and pursuing a path of integrated development. Empowering nuclear energy development with AI will undoubtedly become a new driving force and new business model for the transformation and upgrading of nuclear energy.1. Positive Development Trends in AI Applications in the Nuclear Energy SectorAI technology is increasingly being applied in the nuclear energy sector, especially in nuclear power plants (NPPs), where AI applications have been widely introduced. The collaborative development of nuclear energy and AI technologies has resulted in significant cost reduction and efficiency enhancement. Internationally, AI applications primarily focus on the operation and maintenance of nuclear power plants, nuclear fuel cycle management, safety monitoring of nuclear facilities, and intelligent inspection and robotics applications. Through real-time data analysis, anomaly pattern recognition, and data visualization techniques, AI can significantly improve the construction and operational efficiency of nuclear power, reduce costs, and enhance operational safety. To promote nuclear energy development and the application of new technologies such as AI, the U.S. Nuclear Regulatory Commission (NRC) released the “Artificial Intelligence Strategic Plan: FY 2023-2027” (NUREG-2261), which involves the review and approval of license applications using AI technologies to further promote the development and application of AI in the nuclear energy sector.Currently, domestic nuclear power enterprises are actively empowering nuclear energy safety development by introducing advanced technologies such as AI. For instance, intelligent human-machine interface systems, big data life prediction diagnostics, and operations support systems have been applied in nuclear power plants and uranium mines, primarily based on AI, big data, the Internet of Things, and cloud technology, emphasizing deep integration of technology and human involvement, allowing machines to replace or assist humans in analysis, judgment, and decision-making. Additionally, ongoing projects such as the development and application of autonomous monitoring systems for intelligent nuclear power plants are based on data analysis, theoretical modeling, and experimental design, as well as monitoring, maintenance, and optimization of reactors, helping to predict and resolve potential issues in nuclear energy construction and operation. These R&D projects mainly establish AI-based safety assessment methods to support operators in making decisions under normal and accident conditions, consider human factors in AI application interfaces, and develop adaptive intelligent technologies and software, autonomous operation control systems, etc., to enhance safety and operational efficiency while reducing operational costs.The intelligent nuclear power simulation platform (RINSIM2.0) and nuclear energy modeling and simulation platform (NIMOS) developed by China National Nuclear Corporation can be applied to design verification, operator training, emergency drills, and V&V verification, showing good prospects for promotion. China General Nuclear Power Corporation (CGN) and Zhejiang University have collaborated to develop the “Cloud Book of Nuclear Industry” large language model platform, which already has functions such as knowledge base management, expert systems, and human factor prediction, planning to be put into production application by 2025. The large model “Longyin·Wanjie” jointly developed by China National Nuclear Corporation’s Eighth Institute and Shanghai Artificial Intelligence Laboratory is the first digital productivity platform in the nuclear field in China, integrating the development, use, and management of large model intelligent agents, providing overall solutions for computing power, nuclear field large models, intelligent agents, and AI-native applications. Furthermore, the development and application of new technology products such as small modular reactors (SMRs) also benefit from AI applications in data analysis, simulation, and optimization, promoting the integrated development of AI and nuclear science engineering simulation.To actively promote the comprehensive utilization of nuclear energy, projects like the “Guohe No. 1+” intelligent nuclear energy comprehensive utilization demonstration project and the “Heqi No. 1” nuclear energy heating project are being strengthened under the guidance and coordination of the national directive to build an intelligent energy system, actively drawing on and interacting with various aspects of China’s “Next Generation Artificial Intelligence” major projects, and creating an innovative development model of “Nuclear Energy+” that integrates nuclear energy heating, seawater desalination, nuclear wind and solar storage comprehensive intelligent energy systems, and intelligent multi-energy management platforms, further promoting the multi-field application of nuclear energy.Overall, AI plays an increasingly important role in reducing costs and increasing efficiency in nuclear engineering construction, improving equipment operation and maintenance quality and efficiency, lowering the lifecycle management costs of nuclear power, and enhancing the functionality of nuclear energy development and utilization. However, the application of AI in the nuclear energy sector also faces some risks and challenges, including risks related to digital technology applications such as data security, system failures, human errors, and regulatory challenges, as well as unique issues associated with AI applications, involving the integration of AI with the development and application of nuclear science and engineering technology, optimizing the utilization of AI model resources, and inadequacies in safety regulation and policy frameworks. In the nuclear energy sector, the application of AI technologies must be approached with caution, ensuring information security, reliable decision-making, and compliance, to avoid risks posed by human factors and regulatory obstacles.2. Risks of AI Development and Application in the Nuclear Energy Sector Cannot Be IgnoredThe “Bletchley Declaration” signed at the first Global AI Safety Summit in 2023 emphasized the urgency of understanding and addressing AI safety issues. In 2024, the EU released the “AI Act,” proposing a comprehensive AI tiered regulatory mechanism based on a risk framework. The U.S. has also issued executive orders requiring assessments of AI safety risks. The NRC has collaborated with the Southwest Research Institute to analyze loopholes in regulations related to the use of AI, believing that AI systems are highly likely to produce abnormal outputs, necessitating further exploration of protective measures against safety risks.(1) The Impact of Data Transmission and Processing May IncreaseData is the foundation of AI, and data generation is key to ensuring that AI systems continuously receive new information. However, the decisions made by AI systems may be influenced by data biases, leading to model biases and fairness issues, and in some cases, errors or unexplainable decision processes may arise due to data issues, potentially resulting in unpredictable outcomes. This risk of unexplainability and loss of control can negatively impact the reliability and credibility of decisions. The risk of “data poisoning” cannot be ignored; if “poisoning” affects traditional nuclear systems, it could lead to nuclear safety incidents. If it reaches public opinion, some negative information could more covertly influence the public’s correct understanding.As AI applications proliferate in the nuclear energy sector, intelligent devices such as intelligent monitoring systems and data analysis platforms will collect large amounts of sensitive data, including the operational status and maintenance records of nuclear facilities. If this data is accessed by unauthorized personnel, it could pose safety hazards and operational risks to nuclear facilities. Additionally, AI systems in the nuclear energy sector need to process and analyze vast amounts of real-time data; any anomalies or errors in the data could lead to erroneous decisions, thereby affecting the safe operation of nuclear facilities. Moreover, AI models and algorithms are vulnerable to attacks and theft, which could compromise the construction and operational data and safety strategies of nuclear-related systems, thereby impacting the safety and stability of nuclear energy. Finally, the use of AI technologies in the nuclear energy sector may also involve sensitive data such as unit and personal information. If such information is mishandled, it could lead to privacy breaches and information security issues.(2) The Security Assessment of AI Systems Needs StrengtheningThe construction of AI systems in the nuclear energy sector may involve intelligent maintenance and fault prediction, intelligent device management, intelligent decision support for nuclear reactors, and intelligent safety management, assisting in upgrading traditional nuclear systems. By applying AI technologies and algorithms to real-time data monitoring, scenario simulation, and integrating IoT and big data technologies, it optimizes various stages of nuclear technology development and application, enhancing the safety of nuclear facility construction and operation, energy production efficiency, and waste management capabilities, while also addressing compliance issues. However, AI systems in the nuclear energy sector are typically more complex than traditional systems, making failure modes harder to predict and understand. In cases where AI systems are involved in or assist in decision-making, determining responsibility in the event of an accident could become complicated, making it crucial to pay close attention to safety issues in AI systems used in the nuclear energy sector.Additionally, given that AI applications are based on mathematical models rather than physical models, assessing the reliability of AI systems in the nuclear energy sector is even more challenging. Therefore, using AI applications for design optimization, autonomous operation and control, and safety analysis in nuclear power plants requires careful verification. For instance, the development and application of autonomous monitoring systems for intelligent nuclear power plants must continuously explore intelligent soft and hardware technologies that meet the safety operation requirements of nuclear construction in terms of hardware and system architecture, data processing, safety, and efficiency improvement, gradually establishing a technical verification method and standard system, and iteratively developing in practical applications to achieve high reliability in intelligent systems. Particularly, due to the complexity of the algorithms used in AI applications, which involve analyzing large amounts of data, traditional software verification and confirmation methods are insufficient to assess the reliability of AI applications, and proving that AI-based applications do not interfere with safety functions is challenging.Moreover, due to the limited experience accumulated in safety assessments of AI systems, countries like the United States and other Western nations lack specific guidelines for evaluating AI applications in design and computation, while China’s regulations do not clearly specify whether execution plans must be carried out by humans or AI.(3) Regulatory Standards for the Development and Utilization of Large Models Need to Be StrengthenedCurrently, large AI models are emerging rapidly and have become a significant force in the global AI field. Multimodal large models not only excel in text generation but have also made significant progress in processing and generating audio, images, and other modalities. These technological innovations lay a solid foundation for the application of large AI models in more fields. In recent years, generative AI large models, represented by generative pre-trained large models, have developed rapidly and have been actively applied in new applications empowered by large models, but they also bring unprecedented safety risks.The algorithms of large models may have flaws, and when processing data from the nuclear energy industry, they may produce uncontrollable results due to incomplete or biased data. Some algorithms may have poor interpretability, making it difficult to trace and explain the decision-making process of the model, especially when intelligent agents have autonomous decision-making capabilities; if they lose control, they may lead to unpredictable consequences for nuclear facilities. Additionally, malicious actors may exploit AI algorithms to launch advanced, targeted attacks, covertly sabotaging AI systems and their algorithms. The risks associated with algorithmic models include model poisoning, data poisoning, backdoor implantation, adversarial attacks, and more.Overall, the current global regulatory framework for AI is still in its early stages, lacking a unified and effective coordination mechanism. Regulatory policies vary significantly between countries and regions, making it difficult to comprehensively and timely regulate the application of AI in the nuclear energy sector. Furthermore, the application of digital technologies and AI requires specialized technical support and maintenance teams. Operators may place excessive trust in AI systems, even remaining unaware of the need for human intervention when the system provides erroneous advice. Over-reliance on AI may lead to skill degradation among operators, affecting their responsiveness in emergencies.3. Thoughts and Suggestions for Safely and Orderly Advancing AI Development and Applications in the Nuclear Energy SectorThe State Council’s “Next Generation Artificial Intelligence Development Plan” clearly states that while vigorously developing AI, it is essential to pay close attention to the potential safety risks and challenges it may bring, strengthen proactive prevention and constraint guidance, minimize risks, and ensure the safe, reliable, and controllable development of AI. The State Council’s Office of the Work Safety Committee has issued the “Three-Year Action Plan for Fundamental Solutions to Safety Production in the Energy and Power System (2024-2026),” which explicitly requires continuous enhancement of safety risk monitoring and early warning systems by integrating AI, big data, IoT, and other technologies, with significant improvements expected by the end of 2026 in the intelligent management capabilities of safety production risks.Therefore, the nuclear energy industry should follow the requirements of the “Global AI Governance Initiative” and related policy documents, integrating the actual situation of AI applications in the nuclear energy sector, balancing AI development with safety, maintaining strategic determination, enhancing strategic confidence, and prudently responding to various risks and challenges. While actively and steadily advancing the development and application of AI in the nuclear energy sector, it is crucial to focus on addressing the risks posed by nuclear energy data security, the safety risks of AI systems, and the risks of algorithmic models being attacked.(1) Strengthening Risk Prevention Related to Data SecurityThe development and application of AI in the nuclear energy sector should fully utilize data resources and advanced technologies, adopting data-driven technological innovations and proactive data generation strategies to reinforce data-driven technology iteration and upgrade, promoting innovation and development in nuclear energy. At the same time, it is crucial to pay close attention to data quality and security issues. To mitigate the risks associated with data security, it is essential to actively ensure the collection and application of AI data in the nuclear energy industry.First, it is necessary to ensure that the content does not involve illegal or harmful information. Establish a content review mechanism to prevent the generation of false information and infringing content, and in data transaction agreements, require the provider to guarantee the legal source and handling of data, avoiding legal disputes arising from content generation;Second, conduct data labeling and cleaning to ensure the transparency, interpretability, and fairness of data during the model training stage. This includes formulating labeling guidelines, conducting quality assessments of AI data labeling in the nuclear energy sector, and sampling verification, as well as effectively managing the relationship between simulation data and actual data;Third, establish a stringent data management mechanism, clarify data availability restrictions, and enhance research and innovation on detection and defense methods against “data poisoning,” adopting trustworthy AI technologies such as federated learning, and through modular design, strengthen optimization, diversification, redundancy, and isolation of systems, as well as control and monitoring of software inputs and outputs to avoid adverse impacts;Fourth, establish a user personal information protection system, raising employee awareness of the importance of data security, strictly adhering to relevant regulations, and ensuring user rights to information and choice;Fifth, improve nuclear safety regulation and emergency mechanisms, enhancing safety regulation for AI development and applications in the nuclear energy sector, and establishing a sound emergency response mechanism for data security incidents, strengthening prevention against data breaches, equipment failures, and cyberattacks, and addressing the influence of negative factors.(2) Strengthening Proactive Defense Against Security Risks of AI SystemsWith the widespread application of AI technology, technological advancements have facilitated the integration of nuclear energy and AI systems, while the optimization of AI systems has also promoted advancements in nuclear technology. AI security encompasses data security, algorithm security, system security, network security, and other aspects, aimed at ensuring the safety, reliability, and controllability of AI systems. The AI technology application systems in the nuclear energy sector should strengthen security measures, including both passive and proactive defenses. In principle, AI systems should be isolated from other nuclear energy systems to reduce security risks, with greater emphasis on predicting and identifying potential security threats through continuous monitoring and analysis of system behavior to promptly detect and respond to abnormal activities.In the security research and development of AI systems in the nuclear energy sector, it is recommended to focus on adaptive and intelligent security protection systems, innovations in privacy protection technologies, enhancements in adversarial attack and defense technologies, the establishment of AI security assessment and certification systems, and cross-domain cooperation and technological innovation. To enhance the security of AI systems, the nuclear energy sector can strengthen collaboration with specialized enterprises outside the nuclear energy industry to jointly research and implement comprehensive response plans for AI technology applications in the nuclear energy sector, including the AI security framework for the nuclear energy sector, AI security solutions based on the security framework, AI assessment services, and security detection tools, helping nuclear enterprises anticipate risks and take preventive measures.It is essential to strengthen feedback on the safe operation of AI technology applications in the nuclear energy sector. Based on experience feedback, gradually establish specialized guidelines for assessing AI systems in the nuclear energy sector concerning design, computation, construction, and operation, while emphasizing integrated workflows and embedded system development to address complex and variable working environments. Simultaneously, relevant regulatory standards should be formulated to clarify whether the execution plans of systems are to be carried out by humans or AI.To ensure the safety of AI systems, it is crucial to restrict software inputs and outputs controlled by traditional systems. First, ensure that AI systems in the nuclear energy sector operate in isolated environments during training, testing, and formal deployment phases; second, impose technical restrictions on AI and enhance monitoring to ensure the effectiveness of monitoring systems; third, encrypt and protect the code and parameters of AI systems to prevent theft or misuse by hackers; fourth, apply AI technologies at the modular level or divide systems into smaller, independent modules with “clear functions” to contain any issues or errors within the problem module, avoiding adverse impacts on the overall functionality of the system; fifth, maintain continuous monitoring of AI systems to prevent nuclear equipment failures and maintain system integrity while detecting and responding to potential anomalies.Ensuring the security of AI technology applications requires the joint efforts of society as a whole. It is recommended that national security departments continue to leverage their capabilities to ensure the safety of AI systems in the nuclear energy sector, implementing “diversified, redundant, and isolated systems” to minimize accidental actions and actively utilize AI for security protection. Relevant regulatory agencies should maintain ongoing attention to potential risks and impacts of AI systems in the nuclear energy sector, providing timely support in terms of systems and regulations. Enterprises and institutions in the nuclear energy sector should strengthen cooperation, promptly research and deploy AI security frameworks and unified solutions, as well as AI security assessment services and detection tools, and leverage AI to promote security upgrades. Users in the nuclear energy sector should also update their security knowledge while trying out the latest AI applications, developing good security habits.(3) Coordinating and Optimizing the Selection of AI Large Models and Safe, Efficient Development and UtilizationCurrently, large models are rapidly becoming the foundational base of the intelligent era, widely applied in automatic control of drones, robots, and other fields, serving as the “nerve center” for various physical systems, including industrial control devices. As the demand for large model projects continues to grow, various open-source frameworks have emerged. These frameworks greatly enhance development efficiency and lower the barriers to building AI applications, while also opening new attack surfaces, increasing the security risks of systems or applications. In the intelligent era, where the infrastructure attributes of large models are becoming increasingly prominent, the security of large models is crucial to ensuring the healthy and rapid development of technologies and industries empowered by large models.At present, large model security has become a global consensus and an important strategic high ground in technological competition among major countries. China has released the “Interim Measures for the Management of Generative AI Services,” proposing a regulatory principle that combines innovation and legal governance to support the safe development of generative AI large models. In the nuclear energy sector, firstly, when selecting large AI models for applications such as intelligent review, knowledge base management, intelligent device management, smart operations support, digital twins, virtual reality, and intelligent safety management, it is essential to choose models that are secure and economically viable, such as the DeepSeek large model, which can reduce computing power requirements through dynamic sparse computing, hierarchical MoE architecture, and quantum entanglement-inspired parameter sharing technology, and is compatible with autonomous controllable computing power chips and other hardware, while processing complex tasks related to nuclear energy research, production, and management. These large models must not only match business scenarios but also possess good data quality and strong interpretability to achieve high parameters, high performance, and safety requirements, while also conserving computing resources by optimizing hardware and software use, based on self-controllable chip intelligent computing centers, ensuring safe and efficient performance while reducing overall costs.It is suggested that AI large models in the nuclear energy sector meet the following requirements: first, they must possess strong learning and representation capabilities through extensive data training and deep learning, ensuring data security and system security during the use of large models. Second, they should enhance the overall utilization rate of clusters through model optimization, parameter sharing, parallel strategies, network structure optimization, effective training rate enhancement, and dynamic resource allocation, accelerating the trend of “cost reduction” in the large model industry. Third, they should continuously enrich and develop standardized professional corpuses for large models, improving model performance and generalization capabilities, promoting technological innovation to address challenges such as insufficient knowledge management, excessive low-level labor, and the need to enhance safety analysis capabilities in nuclear energy development. Fourth, they should be applicable in various aspects such as intelligent review, safety assessment, fault diagnosis and prediction, optimized operation and management, and auxiliary design and simulation, ensuring that their research and application are verifiable, safe, and controllable, capable of undergoing regulatory supervision, and continuously enhancing the overall efficiency and safety of the nuclear energy sector, promoting responsible and innovative development of nuclear energy.Secondly, it is crucial to strengthen measures to respond to the risks of large models being attacked. By employing robust training-based and data augmentation-based defense methods, enhance the model’s resilience to interference, and regularly update and retrain models to effectively counter adversarial attacks. Additionally, establish a comprehensive security detection and supervision mechanism to strengthen regulation of the development and use of large models, promptly identifying and addressing potential malicious code and attack behaviors.AI and large models inherently come with security risks, and current research and attention to potential impacts remain insufficient. It is necessary to formulate relevant legal regulations and ethical norms, clarify the scope of use and responsible parties for large models, increase investment to seize the high ground of AI security technology, break through key technologies for large model security, and reasonably plan resource allocation for large models, including selectively shutting down certain technologies for independently developed large models when necessary. At the same time, accelerate the research and development of large-scale cluster technologies and strengthen their flexibility and scalability to support the orderly and clustered development of AI security. Furthermore, given the security threats faced by large models, it is essential to explore continuously at the model, framework, and application levels. The AI ecosystem, supported by large models, has tremendous development potential, and while empowering AI with more capabilities, more effort should also be invested in the security of AI to ensure the entire system is trustworthy, reliable, and controllable.The world today is in a new era of AI development, and nuclear energy must contribute to the popularization and application of AI technologies in China. Therefore, it is recommended that the “Next Generation Artificial Intelligence” special project strengthen support for research and development of “Intelligent Nuclear Energy Systems,” while implementing industry-education integration strategies, and encouraging the nuclear energy sector to establish cross-industry communities to actively participate in the formulation of relevant international and domestic rules, improve the nuclear safety governance system, optimize operational standards for the development and application of AI in the nuclear energy sector, and provide professional training for relevant personnel, continuously growing the practical and compound talent pool in the field of “AI+Nuclear Safety,” promoting the construction of the national intelligent energy system, and playing a crucial role in ensuring energy security transformation, achieving dual carbon goals, leading industrial technological transformation, and promoting the development of new quality productivity.
