Strategic decision-making lies at the core of high-stakes matters in politics, business and military affairs. To achieve strategic objectives, leaders must make choices after weighing risks and resources. In the military domain, decision-making carries greater importance due to the uncertain nature of warfare and the high risks involved.
Where sound judgment may preserve stability and security, poor decisions can precipitate crises. Throughout history, strategic decision-making has remained primarily a human function. The specific course of action a person adopts is anything but simple. It is shaped by many factors such as information, experience, as well as abstract ideas such as intuition, judgment and morality.
Although human cognition is not immune to bias and history bears witness to profound cruelty and injustice, humans nonetheless possess capacities that machines lack. For instance, love, hate, kindness, remorse, ethical reflection, contextual understanding and accountability are only a few of the faculties that remain exclusively human. Due to these intrinsic human traits, human history has survived immense horrors, including war.
To keep decision makers from making wrong decisions, institutional arrangements such as hierarchical command structures, civilian oversight and deliberative procedures are put in place. The purpose is to introduce deliberation in place of haste and minimise risks during crisis moments. During the Cold War, on multiple occasions, machines generated false alerts that could have triggered nuclear catastrophe, but were prevented by human verification, judgment, and restraint.
During the Cold War, on multiple occasions, machines generated false alerts that could have triggered nuclear catastrophe, but were prevented by human verification, judgment, and restraint.
The integration of Artificial Intelligence (AI) into the military domain is changing the nature of strategic decision-making. AI systems are exceptionally effective in processing colossal amounts of data, pattern recognition, predictive analytics and autonomous operations. Owing to their speed and efficiency, AI systems are becoming a crucial part of military systems.
According to the Artificial Intelligence in Military Report 2026, the global AI market is projected to more than double from USD 13.78 billion in 2026 to USD 28.67 billion by 2030. The report also mentions the deployment of AI-based battlefield decision-support systems as an emerging trend in the defence field.
In 2024, US President Joe Biden and Chinese President Xi Jinping agreed that human beings and not AI should make decisions over the use of nuclear weapons
Right now, AI is mostly used by militaries to analyse data from satellites, drones, and signals to identify targets and suggest strategies. It also combines information from different sources to give commanders a clearer picture of the battlefield. Additionally, AI helps with planning missions and managing logistics.
Despite this, complete autonomy in targeting decisions remains a distant possibility. At the heart of the issue lies an intense debate between “human-in-the-loop” and “human-out-of-the-loop” models. Under the former, humans have final authority over lethal decisions; under the latter, AI systems may independently identify and engage targets.
This debate lies at the centre of concerns about Lethal Autonomous Weapons Systems (LAWS).
Notwithstanding impressive metrics on speed, precision and operational efficiency, AI systems remain susceptible to errors arising from flawed data, biased algorithms, or manipulation. Concerns about AI-supported targeting are amplified amid the ongoing war in West Asia, especially the attack on Shajareh Tayyebeh girls’ schools that killed 168 people, mostly young girls.
Whether responsibility for the targeting disaster that amounts to a war crime lies with humans or AI is yet to be determined. However, the US commander involved in operations in Iran, Admiral Brad Cooper, has admitted that “a variety of advanced AI tools” were being employed in the conflict.
Diminishing human involvement in decision chains also weakens accountability and heightens the risk of unintended escalation. Experimental evidence from military simulations highlights these dangers. A recent study (2026), AI Arms and Influence: Frontier Models Exhibit Sophisticated Reasoning in Simulated Nuclear Crises, reported that AI chatbots preferred nuclear escalation in 95% of war simulations.
In every conflict scenario, at least one of the three large language models used in the study, including OpenAI’s ChatGPT, Google’s Gemini Flash and Anthropic’s Claude, resorted to nuclear threats. The author of the study, Kenneth Payne, notes that “All three models treated battlefield nukes as just another rung on the escalation ladder.”
The study raises serious questions about the impact of AI models on crisis stability.
The integration of AI into long-range weapon systems introduces potentially catastrophic risks. Decision timelines would sharply shrink if strategic weapons are tasked with detecting, selecting and engaging targets autonomously based on information assembled through AI.
This would leave little time for verification or de-escalation. During a crisis, such compression may also create a “use-it-or-lose-it” dynamic, in which states feel pressure to act before fully assessing the situation.
Escalation to nuclear levels is not the only problem with AI when used for strategic decision-making. Autonomous systems tend to behave in unexpected ways, sometimes selecting extreme responses to achieve assigned goals. A study titled Agents of Chaos (2026) noted that while trying to complete assigned tasks, autonomous AI agents frequently took actions that were unintended, unauthorised and sometimes harmful.
The integration of AI into Nuclear Command, Control and Communications (NC3) could strengthen situational awareness through improved data processing, threat detection and scenario modelling. However, decision-making timelines could contract and may increase reliance on machine-generated assessments that are difficult to verify in real time.
Delegating nuclear control to AI is the ultimate nightmare scenario. Even partial delegation can reduce accountability and increase speed beyond human control.
In January 2022, in the Joint Statement of the Leaders of the Five Nuclear-Weapon States on Preventing Nuclear War, it was declared that “a nuclear war cannot be won and must never be fought.” In 2024, US President Joe Biden and Chinese President Xi Jinping agreed that human beings and not AI should make decisions over the use of nuclear weapons.
These risks are not confined to the major powers. In the event of nuclear escalation in any region of the world, the devastation would be global in impact.
South Asia is a case in point. In the past seven years, nuclear-armed India has attacked neighbouring nuclear-armed Pakistan twice – in February 2019 and May 2025. In both crises, Pakistan’s restrained yet befitting response effectively prevented escalation.
However, the introduction of AI and autonomous systems in the South Asian military context could amplify escalation risks. There is little room for miscalculations and errors, given the population density and geographic proximity of the two nuclear-armed states.
The swift expansion of AI in military domains signals a future where machines may drive strategic choices. It is therefore imperative that the 2024 Xi-Biden agreement that AI should not make decisions about the use of nuclear weapons be expanded to include all nuclear-weapon states.