Experts See Risk and Reward to Integrating AI in Nuclear Weapons

Executive Summary:

• Chinese experts see potential in the ability of cyberattacks enabled by artificial intelligence (AI) to disrupt nuclear infrastructure and increase the pressure to use nuclear weapons in a crisis.
• The development of early warning capabilities toward a launch-on-warning posture increases Beijing’s impetus to integrate AI into data processing to inform decisions over nuclear use.
• There is significant ongoing debate on the threats AI-enabled conventional threats pose to the People’s Liberation Army’s nuclear forces, the effectiveness of remote-sensing in undersea warfare, and the vulnerabilities of using AI to process early warning data and generate options for decision-makers to respond to nuclear threats.

In April 2025, Zhang Gaosheng (张高胜), a researcher at the China Institute of International Studies, penned an article in The Paper pinpointing several mechanisms in which the integration of artificial intelligence (AI) into nuclear command, control, and communications (NC3) systems will increase the risk of nuclear escalation (The Paper, April 11, 2925). In particular, as AI technologies become increasingly embedded in critical nuclear infrastructure, the potential for miscalculation, system vulnerabilities, and unintended escalation grows more acute. It is, therefore, critical to understand how the PRC plans to integrate AI into its nuclear strategy.

Although Biden and Xi previously agreed on the need to maintain human control over the decision to use nuclear weapons, the decision to integrate AI into nuclear command, control, and communications NC3 is far more complex than the decision to maintain a human-in-the-loop (Ministry of Foreign Affairs, November 17, 2024). From processing early-warning data to autonomous targeting, there are numerous ways of integrating AI into NC3 while maintaining people within the decision-making process. As the performance of AI models rapidly improves, Chinese experts have identified opportunities to incorporate AI in NC3. They have also discussed the weaknesses of doing so and the risks AI may pose to the survivability of its nuclear forces.

AI-Enabled Cyber and Conventional Systems Threaten Nuclear Survivability

Chinese experts are broadly concerned about AI’s threat to nuclear strategic stability. The ability of AI to detect targets, sort through data, and improve decision-making speed can enhance counterforce capabilities and provide offensive advantages that can threaten adversary retaliatory capabilities (Center for International Security and Strategy, September 2, 2019). Chinese analysts have identified U.S. decisions to integrate AI into components of its NC3 system as threatening PRC second-strike capabilities. These decisions to enhance counterforce capabilities create arms race dynamics and drive up the requirements for a “strong strategic deterrent system” (强大战略威慑力量体系). This, in turn, will trigger Chinese countermeasures to upgrade the mobility and penetration capabilities of its nuclear forces (Shanghai Institutes for International Studies, March 2025).

The Cyber-AI-Nuclear Nexus

PRC writings have emphasized emerging threats from AI-enhanced cyberattacks. As Xi Jinping declared while addressing the Central Leading Group for Cybersecurity and Informatization in 2014, “Without cybersecurity, there is no national security” (没有网络安全就没有国家安全) (Xinhua, February 27, 2014).

The development of AI-enabled cyberattacks could significantly increase the vulnerability of PRC nuclear forces. AI-powered autonomous agents and advanced persistent threats (APT) can analyze vast amounts of data to systematically detect, analyze, and exploit weaknesses in target systems. AI-enabled malware can also automatically alter its code to evade detection and recognize where potential zero-day vulnerabilities may appear. Chinese analysts have observed that such developments in AI-enabled cyberattacks could more easily allow adversaries to find and exploit weaknesses in systems that could disrupt nuclear infrastructure. The proliferation of AI and cyber capabilities could enable third parties and terrorists to sow discord among nuclear-armed countries through “smart intrusions” (智能入侵).

The speed at which AI-enabled cyberattacks can “paralyze” (瘫痪) nuclear weapons systems could increase pressures to use nuclear weapons in a crisis, according to analysts from the National University of Defense Technology. This is due to the difficulty of attributing responsibility, detection, and issuing warnings in a short time period—especially if the country maintains a launch-on-warning posture (Information Security and Communications Privacy, October 11, 2021).

AI can also improve cyber defenses by autonomously conducting vulnerability analysis, searching for breaches, and identifying APTs. Some speculate that AI can intensify a measure-countermeasure race in which several actors repeatedly identify vulnerabilities in adversary nuclear weapons systems and take corresponding actions to rectify vulnerabilities in their own systems (Center for Canadian Studies [CCS], November 24, 2020; Information Security and Communications Privacy, October 11, 2021).

Such vulnerabilities make the PRC’s nuclear infrastructure a likely target in any crisis or wartime environment that involves cyber. Experts point to the Bush and Obama administrations’ efforts to use cyberattacks to sabotage North Korea’s nuclear weapons program as indicative of scenarios the PRC may face (CCS, November 24, 2020). To address the threats posed by AI-enabled cyberwarfare, analysts writing in the Journal of Intelligence recommend increasing redundancies in information network systems, enhancing tailored defenses of infrastructure, and making the structure of critical infrastructure less predictable and harder to map (Journal of Intelligence, 2021).

Blurring Conventional and Nuclear Threats

AI can also enhance conventional threats to nuclear forces. PRC experts have long feared that “non-nuclear strategic weapons” (非核战略武器) could threaten the country’s nuclear forces. As a result of transformative technological advances in precision guidance and information networks after the Cold War, National Defense University scholar Zhang Yan (张岩) emphasizes that conventional weapons and long-range precision strike capabilities could be used to threaten an adversary’s nuclear forces (Military History, March 2018). Another analyst notes that the enhanced ability of the United States to detect, track, and attack missiles through precision-guidance threatens PRC nuclear capabilities and weakens mutual vulnerability (Journal of International Security Studies, March 29, 2019).

The integration of AI with conventional weapons creates additional means of threatening PRC nuclear forces. As the United States and the PRC both attempt to integrate autonomy into drones to enhance conventional capabilities, experts have indicated that drone swarms can be used to threaten PRC nuclear facilities. As an op-ed in the PLA Daily, the official mouthpiece of the People’s Liberation Army (PLA), highlights, drone swarms can be used to penetrate multi-layered air defenses and launch pre-emptive strikes that threaten the PRC’s nuclear counterattack capabilities and allow adversaries to gain advantage via “unilaterally assured destruction” (单方面确保摧毁). These drone swarms can be used to target nuclear delivery vehicles, NC3 infrastructure, and early warning systems used to enable nuclear retaliation (PLA Daily, July 20, 2021). [1]

Some experts posit that AI will not enhance the ability of conventional forces, especially drone swarms, to threaten an adversary’s nuclear delivery systems. Drones carrying conventional payloads are not effective in destroying reinforced silos and are less ranged compared to traditional ballistic missiles. Geographic factors also limit the ability of drones and other autonomous vehicles to hunt ballistic missile transporter erector launchers or nuclear submarines (SSBNs) located in vast oceans, mountain caves, or other remote locations (The Journal of International Studies, 2020).

AI Opportunities in Early Warning and Remote Sensing

Integrating AI in Early Warning and Data Processing

Since 1964, the PRC has maintained a no-first-use declaratory policy, committing to not be the first country to use nuclear weapons in any conflict under any circumstances (Ministry of Foreign Affairs, July 23, 2024). At the same time, however, the U.S. Department of Defense assesses that the PRC is building a diverse array of space and ground-based early warning sensors designed to detect nuclear launch and provide the PRC an option to maintain a launch-on-warning posture and rapidly retaliate before a nuclear weapon has been detonated (Department of Defense, December 18, 2024).

Using early warning systems to inform decisions over nuclear use is data-intensive and labor-intensive. Chinese experts highlight that commanders and decision-makers may be overwhelmed by the large amount of information about adversary actions, preventing commanders from making quick and informed decisions. AI integration in nuclear early warning systems can quickly screen incoming data, improve the quality of data, and distinguish between false and real warnings to give decision-makers more time to respond. Outside of data processing, scholars foresee AI being used in a generative role to provide commanders with potential options to respond to nuclear and non-nuclear threats (Journal of Intelligence, 2021).

Some experts nevertheless fear that AI models used for decision support in NC3 can be deceived, manipulated, or produce erroneous judgments in early warning systems. One Chinese observer asserts that a lack of high-quality training data could hinder the effectiveness of AI models, especially in the nuclear domain. With fewer relevant data points and limited access to data, AI models are likely to produce unreliable outputs that could lead to failures in sensitive nuclear weapons systems. More maliciously, actors can use AI to create deepfakes that simulate videos or audio of senior civilian and military leaders that are leaked to another country’s intelligence collection and analysis organization. Experts point out that such deepfakes can contribute to creating crisis instability, misperception, and strategic misjudgment (Information Security and Communications Privacy, October 11, 2021). AI models may also be susceptible to data poisoning by nefarious actors. This can be done when adversaries deliberately expose or hide information based on how an AI model weighs information to create analysis or predictions, causing the model to draw incorrect conclusions (PLA Daily, July 20, 2021).

Stalking the Sea

The United States relies heavily on ballistic missile submarines (SSBNs) as a key component of its nuclear triad to maintain its nuclear deterrence. It maintains 14 Ohio-class SSBNs—a larger fleet than the PRC’s six operational Jin-class SSBNs, which it deploys for strategic deterrence missions. (The sea leg of the nuclear triad is one of the most survivable and critical components of the nuclear triad due to the difficulty of tracking SSBN in the deep sea.)

Chinese analysts indicate that AI technology and remote sensing can enable the deployment of autonomous underwater unmanned vehicles (UUVs) to detect, track, and attack SSBNs to reduce their survivability and limit their range. Their relatively cheap production could allow the PRC to deploy UUVs in large numbers to monitor choke points that SSBNs must pass through to reach or leave their patrol areas. Once the SSBN is identified, AI and remote sensing technologies could enable UUVs to continuously track SSBNs and prevent them from entering specific areas. One Chinese expert from the University of International Relations in Beijing points out that researchers are investigating the effects AI could have on SSBN survivability, because the PRC maintains fewer SSBNs that all perform worse than those of other nuclear powers (Theory and Practice of International Relations, December 17, 2024). Others, however, dismiss the effect AI has on undersea warfare, pointing to the limitations on the ability to deploy enough UUVs to cover a wide enough stretch of sea to track and destroy SSBNs (The Journal of International Studies, 2020).

Conclusion

The question of integration of AI in its nuclear forces continues to provoke active and ongoing debate within the PRC. Experts perceive AI as both an enabler and a threat in the nuclear domain. While it can enhance early warning, data processing, and decision support, it also introduces new vulnerabilities to nuclear forces. Scholars are especially wary of AI-enabled cyber threats to nuclear infrastructure and the potential vulnerabilities of AI models to cyberattacks in NC3. Understanding how the PRC conceptualizes the integration of AI into its nuclear strategy is critical for contingency planning, crisis management, and assessing future risks to strategic stability, as it allows planners to anticipate how the PRC may respond to U.S. technological initiatives.

Across authoritative releases and white papers, the PRC for years has been emphatic about AI’s potential in areas ranging from military systems to economic development. It has been opaque, however, on the nature of its nuclear build-up and operational concepts, let alone on specific plans for integrating AI within nuclear weapons systems. This means that any such analysis inherently has limited predictive power. Outlining a range of potential applications for AI in nuclear weapons as identified by Chinese experts nevertheless remains a valuable exercise—their views could inform future PRC policy and military development. Academic journals and PLA analyses likely will continue to publish on the topic, and are worth paying attention to as the debate develops in the years ahead.

Notes

[1] Zhang Huang and Du Yanyun. “The Trend of Militarizing Artificial Intelligence and its Impact on Security” (人工智能军事化发展态势及其安全影响). Foreign Affairs Review (外交学院学报), no. 3 (2022): 99-130. DOI: 10.13569/j.cnki.far.2022.03.099.