Daniel Craig's Skyfall, a film that challenged traditional James Bond tropes, features a pivotal conversation between Bond and Q. Q boasts that he can cause more damage from his laptop in his pyjamas than Bond could in a year in the field. Bond retorts, 'Every now and then, a trigger has to be pulled.' Q replies, 'Or not pulled… it’s hard to know which in your pyjamas!' This exchange encapsulates the dilemma of nuclear command and control, a dilemma that Stanislav Petrov faced on 26 September 1983.
Stanislav Petrov: The Man Who Saved the World
On that day, Petrov, deputy chief for Combat Algorithms at the Soviet Union’s Serpukhov-15, decided not to pull the trigger. The early warning system indicated a missile fired from Montana heading toward Moscow. Within minutes, more missiles appeared to be heading toward the USSR. Petrov followed his instinct instead of protocol and waited. After a nerve-racking wait, he learned it was a false alarm due to a system malfunction. This decision likely prevented a nuclear war.
The 1979 NORAD Incident
Four years earlier, in 1979, the North American Aerospace Defence (NORAD) headquarters in Colorado and the Strategic Air Command Centre were alerted that Moscow had initiated a large-scale nuclear attack. Washington readied countermeasures and placed bombers and missiles on alert. However, it was soon realized that a technician had erroneously inserted a training tape simulating a Soviet attack into a NORAD computer. Again, human intervention prevented catastrophe.
AI in Modern Nuclear Command and Control
Fast forward to 2026. We live in an age of AI and ever-learning algorithms. AI processes data far faster than human cognition. If confronted with a similar situation, how would an AI-enabled system respond? AI offers advantages like clearer strategic orientation and enhanced data analysis from satellite imagery, radar, signals intelligence, and cyber indicators. It can also strengthen Personnel Reliability Programs through continuous monitoring. However, AI has flaws: data poisoning, indirect prompt injection, cyber intrusion, and adversarial spoofing. These threaten negative control of nuclear weapons and increase the risk of accidental escalation.
The Transparency Paradox and Automation Bias
In a nuclear environment, once an AI-enabled early-warning system flags a high-probability threat, human operators cannot verify it due to time compression and lack of contextual data. This creates a transparency paradox and automation bias—human over-reliance on machine input. While AI processes complex battlefield data in seconds, it deprives leadership of human deliberation and informed choices, pushing toward a launch-on-warning posture. This raises concerns about first-mover advantage and the use-it-or-lose-it dilemma.
The Risk of Accidental Launch
It would be all about pulling the trigger, as the choice of not pulling the trigger through diplomatic de-escalation would be non-existent. AI cannot work around edge cases or identify false alarms. Unlike humans who have identified false alarms, AI would initiate an actual launch. Despite promising developments like the 2024 Biden-Xi affirmation, the 2024 Responsible AI in the Military Domain (REAIM) summit, and a 2025 UN General Assembly resolution on AI risks to nuclear weapons, two points must be considered. First, a broader dialogue involving all stakeholders is needed to develop a unified understanding of risks and opportunities. Second, as long as AI is seen as a strategic tool for dominance, reaching a collective position will be impossible.
The Need for Human Oversight
Lieutenant General Nauman Zakria, commander 1 Corps and head of the Army Rocket Force Command, stressed at the 23rd IISS Shangri-La Dialogue that rapid advances in AI, autonomous systems, cyber capabilities, quantum technologies, and multi-domain operations are transforming military decision-making and strategic competition, while introducing new vulnerabilities, risks of miscalculation, and unintended escalation. As states and critical infrastructures become dependent on interconnected technological ecosystems, the erosion of predictability and compression of decision-making timelines are reshaping conflict and deterrence. It is high time that humanity addresses this issue to avoid being the last generation in control of its affairs.
