The application of artificial intelligence (AI) in nuclear weapon systems is poised to expand in the coming years promising potential benefits but also the risks that could impact strategic stability. The automation of nuclear weapon systems is not a novel idea as its origins can be traced back to the Cold War era. In the Cold War era, both the USSR and the US integrated automated systems into their nuclear weapons, primarily for informed decision-making, but these systems were not fully autonomous. With the advancements in AI, its application in early warning systems and decision support mechanism can have recognizable benefits.  Likewise, its use in nuclear weapon systems can bolster security by defending against possible attacks both physical and cyber. Yet, as AI technology remains in its maturation phase, fully autonomous nuclear weapon systems without human involvement may elevate the risk of decision-making errors, potentially leading to accidental nuclear conflicts.

During the Cold War, both the US and USSR ventured into semi-automated nuclear weapon systems, laying the foundation for AI integration in such systems. A prime example from the Soviet side is the “Perimeter,” also known as “dead hand.” This automated system was developed out of concerns regarding a potential first nuclear strike by the US, designed to serve as a fail-deadly option following a decapitation strike. It was programmed to initiate an automated launch by analyzing radioactive and seismic data, in addition to processing information from early warning systems and satellites. Similarly, in the 1980s, the US envisioned WOPR (War Operation Plan Response), an automated nuclear response system that assessed the likelihood of an enemy nuclear attack and could potentially launch a pre-emptive first strike based solely on assessments. Both cases featured minimal or no human involvement in the nuclear weapon launch process.

Major nuclear-armed states are actively modernizing their nuclear weapon systems, with AI integration emerging as a key avenue for optimizing their capabilities. A primary focus lies in enhancing early warning and reconnaissance systems, where AI can offer comprehensive situational awareness, bolstering defense and confidence in retaliatory capabilities. Notably, the US is transitioning from Cold War-era computers with floppy disks to AI-integrated supercomputers. These supercomputers will analyze and correlate vast volumes of data collected from satellites and detection systems. Advancements in hypersonic missile technology are reducing the time available for leaders to make retaliatory strike decisions, intensifying psychological pressure. Integrating artificial intelligence into the OODA (Observe, Orient, Decide, Act) loop, in tandem with human judgment, can provide crucial time for leadership to authorize launch decisions. The benefits of AI integration in nuclear weapon systems are evident but it is crucial to acknowledge the considerable risks it may pose.

The prospect of a fully autonomous AI-integrated nuclear weapon system evokes chilling concerns, as the associated risks seemingly outweigh the benefits. Mutual vulnerability among nuclear-armed states has long been a cornerstone of deterrence. Speculations about AI-enhanced hypersonic missiles, with their potential for heightened speed, precision, and maneuverability, could render defense against them virtually impossible. The use of AI in Intelligence, Surveillance, and Reconnaissance (ISR) capabilities, extending to the detection of submerged enemy submarines, could destabilize first-strike stability. This might erode states’ confidence in their retaliatory capabilities and spark a new nuclear arms race. Russian President Putin’s mention of reviving a Soviet-era ‘dead hand’ system with AI integration raises alarms, as fully autonomous machines are susceptible to errors. The pushback against automating nuclear weapon systems is a response to the imperative to prevent any accidental nuclear catastrophe.

The Cold War era offers stark reminders of the potential calamities associated with fully automated nuclear weapon systems, amplified by the capacity for machines to err. A notable illustration is the case of Soviet air defense officer Stanislav Petrov, who, when confronted with multiple incoming ballistic missile alerts on radar, chose not to relay the information to higher authorities. Subsequent investigations revealed that there was, in fact, a radar error, erroneously suggesting a potential US first strike on the Soviet Union. In the contemporary context, concerns emerge from the prospect of removing human involvement from the OODA loop. The worry centers on a hypothetical scenario where AI-integrated nuclear weapon systems make a faulty decision, with no human intervention akin to Petrov’s role. Handing over launch decisions entirely to AI raises the specter of inadvertent nuclear escalation.

The burgeoning AI technology presents the potential to assume complex tasks currently performed by humans; however, the integration of AI into nuclear weapon systems, leading to full automation, may paradoxically prove destabilizing rather than stabilizing. Currently, there exists no international law or treaty explicitly prohibiting states from pursuing such autonomous systems. Given the rapid proliferation of this technology, there is an imperative to establish new international laws or arms control agreements to mitigate the adverse implications of this emerging technology. Great powers should engage in earnest negotiations to address the potential ramifications it poses for the contemporary nuclear order. Ensuring that humans remain integral to the decision-making process in nuclear command and control is pivotal in averting inadvertent escalations and preserving global security.