Τελευταία Νέα
Διεθνή

Something will happen on November 19: Why are the 3 major nuclear powers taking Artificial Intelligence out of the equation?

Something will happen on November 19: Why are the 3 major nuclear powers taking Artificial Intelligence out of the equation?
Artificial intelligence does not necessarily need to rebel against humanity. It is enough to convince humans that they can win by destroying each other.

Why did the three major superpowers remove artificial intelligence from their nuclear arsenal?

For centuries, major powers dreamed of a form of supremacy that would allow them to achieve victory without risks or losses. As military means grew increasingly sophisticated, the temptation to believe victory was finally guaranteed grew stronger, especially when such a promise relied not on overconfident strategists, but on precise mathematical calculations. Yet suddenly, the world's most powerful states began discussing whether it was time to place limits on their own power.

A magic wand for nuclear war

Ahead of the scheduled November 19, 2026, meeting between Donald Trump (US), Vladimir Putin (Russia), and Chinese leader Xi Jinping, an unexpected topic is being prepared in China: artificial intelligence as the subject of a joint statement. As the South China Morning Post reported, representatives of the three superpowers spent weeks searching for an issue on which they could reach an agreement. While topics such as Ukraine and the denuclearization of the Korean Peninsula were sidelined, the idea emerged for a joint statement regarding limiting the role of artificial intelligence (AI) in nuclear weapons control. What happened that led these rival powers to discover a mutual interest in this field?

Why is artificial intelligence being taken out of the equation?

Not long ago, artificial intelligence seemed like little more than a fun toy, capable of generating strange images, composing mediocre poetry, or synthesizing knowledge in seconds. Today, it performs these tasks with remarkable proficiency. At the same time, it solves mathematical problems once accessible only to the most brilliant scientists. In 2025, Google's "Gemini Deep Think" system reached the level of gold medal winners at the International Mathematical Olympiad, solving 5 out of 6 problems. In May 2026, an OpenAI model shattered a geometric hypothesis that had puzzled mathematicians for 80 years. And in September 2026, the artificial intelligence model "Claude" took 11 days to translate the proof of French mathematician Pierre de Fermat's Last Theorem into a format a computer could mathematically verify. Humans no longer need to write code themselves; they can simply assign the work to a team of robotic programmers. A massive leap has been achieved in the programming sector. Systems like Codex and Claude Code no longer merely provide suggestions to human programmers; they analyze existing code, design solutions, write software, execute tests, and fix their own bugs. The human programmer no longer writes code personally, but manages a team of robotic programmers—entities that work simultaneously, never tire, and can complete tasks in a few days that humans would need months to finish.

Artificial intelligence and the military sector

The military sector, naturally, did not stay on the sidelines. The American company Palantir developed the Maven system, which integrates data from satellites, drones, radar, and intelligence sources to assist in target identification, operational planning, and decision-making. NATO acquired the system in 2025, and in 2026, the Pentagon set a goal to establish Maven as a core component of military command and control infrastructure. This is no longer science fiction: artificial intelligence (AI) is becoming an integral part of actual operations within military headquarters. Although Maven does not currently control nuclear weapons, it clearly demonstrates new, advanced capabilities to military leadership.

It will detect even the most well-hidden strategic missiles

There is no guarantee that such technologies will not eventually reach a level capable of detecting even the most well-hidden strategic missiles, calculating strikes against them, and simultaneously coordinating missile defense. For military strategic analysts, this is equivalent to a "magic wand." Historically, nuclear weapons functioned as a deterrent guarantee: even after a devastating strike, an adversary could retaliate in a manner that rendered any concept of "victory" completely meaningless. Now, however, a new possibility arises: if artificial intelligence can locate an opponent's missiles, calculate how to destroy them, and intercept any that survive, why not attempt to win a nuclear war?

What artificial intelligence does not know: Errors and indications

A fundamental challenge exists: even the most advanced artificial intelligence would have to search for assets that the adversary has learned to keep hidden for decades, such as nuclear-powered strategic submarines. According to estimates by the Federation of American Scientists (FAS), Russia is projected to possess 13 such submarines by 2026, carrying approximately 704 nuclear warheads on sea-launched missiles. Each Borei-class submarine has the capacity to carry 16 ballistic missiles, each equipped with multiple warheads. The United Kingdom recently announced a 90 million pound investment in "Atlantic Net," a system consisting of marine drones, underwater sensors, and artificial intelligence. This system is designed to collect and analyze data regarding activity in the North Atlantic, locate Russian submarines, and alert NATO ships and submarines tasked with anti-submarine warfare. However, if even a single Russian submarine remains undetected, its 16 missiles would pose a massive problem for NATO.

Land-based missiles an even tougher problem

The challenge is equally difficult regarding land-based missiles. According to the same Federation of American Scientists estimates, Russia possesses approximately 324 intercontinental ballistic missiles, including Yars, Topol-M, and Voevoda systems, which have the ability to move and evade detection by intelligence services. To protect these mobile missile units, Russia utilizes Peresvet laser systems, one of whose functions is to neutralize satellite surveillance. Furthermore, even if NATO intelligence services located a Russian mobile launcher, it would not simply remain in place waiting for an incoming strike. Let us assume, for the sake of argument, that an initial strike would be successful. The issue does not end there: one would have to face a retaliatory strike. What would artificial intelligence have to say about that?

What about the American Pentagon?

What does the American Pentagon possess? The Ground-based Midcourse Defense (GMD) system has 44 ground-based interceptor missiles. However, experts from the US Congress have already pointed out that this system would be unable to repel a massive missile attack from Russia or China. Nuclear war is governed by a unique logic of its own: even if 99% of 1,000 nuclear warheads were intercepted, the 10 remaining would suffice to destroy several major cities. It is entirely possible to win almost every individual conflict and, nevertheless, lose the war. Furthermore, military command and control systems have exhibited errors in the past, mistakes for which humanity could have paid with its very existence.

What happened in September 1983?

In September 1983, the Soviet early warning system "Oko" (Eye) emitted a signal that American missiles had been launched. The duty officer, Stanislav Petrov, questioned the validity of the alarm, realizing the price of retaliation would be excessively high, and reported that it was a false alarm. It was later revealed that satellites had mistaken sunlight reflecting off clouds for exhaust trails from operating rocket engines. In January 1995, Russian radar systems mistook a Norwegian scientific research rocket for a potential hostile missile. The alarm reached President Boris Yeltsin, to whom the "nuclear briefcase" was even brought. Fortunately, Russia's first post-Soviet president was occupied with something more important at that exact moment.

Dangerous for nuclear weapons: Artificial intelligence

In June 2025, the Stockholm International Peace Research Institute (SIPRI) published a study regarding the impact of military artificial intelligence (AI) on the risk of nuclear war. Its conclusion: rather than eliminating vulnerabilities, new technologies may actually increase the vulnerability of strategic arsenals. Although artificial intelligence will undoubtedly continue to evolve by identifying more targets and calculating strike outcomes with increasing accuracy, there is no guarantee an adversary will remain without the means to retaliate, even after an extremely successful initial strike. And the price for such a mistake is not merely a lost battle, but a lost planet.

The historical reality

Yet the fact that victory is not guaranteed does not mean a nuclear war is impossible. After all, the decision to attack is made by humans, and humans are prone to believing all sorts of things. Military history is full of commanders who launched wars, utterly convinced the adversary would be crushed in one or two lightning blows. In 1914, the German General Staff expected to defeat France in six weeks. In 1941, Hitler hoped to conquer the Soviet Union in an equally short timeframe, and both erroneous estimates led to protracted world wars. In 1940, Mussolini planned to crush Greece in 8 days and failed. In 1979, the USSR hoped to quickly bring Afghanistan under its control. In 1982, the Argentine dictatorship never expected Britain to fight for the Falklands. In 2003, George W. Bush declared victory in Iraq just a month after the invasion, yet time and again, reality ironically upended the chessboards of these "great" strategists. In 2001, the American military promised it would quickly finish off the Taliban and then withdraw from Afghanistan. Although the Taliban were indeed dispersed within two months, the actual withdrawal did not happen until 20 years later, at which point those exact same Taliban immediately retook the Afghan capital of Kabul.

Delusional politicians

Today, these delusional politicians and generals living in their own world have acquired a new, hyper-intelligent advisor capable of reinforcing their fantasies with billions of calculations, exceptionally accurate intelligence, and highly persuasive mathematical models. And how could they possibly resist? This issue even has a scientific name: "automation bias," the human tendency to show excessive trust in automated systems.

Artificial intelligence is extremely dangerous for nuclear weapons

In June 2025, the Stockholm International Peace Research Institute (SIPRI) warned that the use of artificial intelligence (AI) in military command and control could increase the risk of nuclear escalation without even granting a machine the authority to deploy weapons. An algorithm could misinterpret an adversary's intentions, suggest a dangerous course of action, or present a doubtful prediction so persuasively that commanders would not bother to question it, especially when only minutes or perhaps seconds remain for decision-making, and alarming indications that look entirely realistic begin appearing on screens: mushroom clouds, thermal engine traces, or something strikingly similar to them. The more complex a system becomes, the harder it is for a human to verify it.

Stanford research causes terror

This has already been proven experimentally. In 2024, researchers at Stanford University conducted war games involving 5 major artificial intelligence models. These systems undertook the management of virtual states, the conduct of negotiations, the formation of alliances, and the making of military decisions. All 5 models exhibited a tendency toward escalating conflicts, and in some scenarios even resorted to the use of nuclear weapons. Furthermore, the artificial intelligence systems justified their decisions by invoking known strategic logic, citing the need to deter the adversary, demonstrate strength, or exploit the advantages of a first strike. Of course, this was merely a simulation. However, in essence, the artificial intelligence used was of the same type that dominant nuclear powers are now enlisting into service.

What experiments teach us

Let us imagine, then, a military command center of the future. An artificial intelligence reports: the enemy clearly intends to launch a first strike, almost all their missiles have been located, and a plan for their destruction has been calculated. Any subsequent retaliation will be intercepted by missile defense systems. The probability of success is 99%. Delay is dangerous, striking first is more advantageous. It does not need to seize power; it is easier to sneak in through the "back door" where no one expects it, not as an invader, but as a well-meaning and intelligent advisor.

What does this mean?

In November 2024, Biden and Xi Jinping agreed that the decision to use nuclear weapons must remain in human hands. Now, according to the South China Morning Post, the US, Russia, and China are discussing ways to limit the use of artificial intelligence (AI) within nuclear systems. However, the fact that the final word belongs to a human provides no guarantee of security. After all, that human makes their decision based on the information they receive. This is the real issue ahead of the meeting between Trump, Putin, and Xi Jinping in China: the top nuclear powers must learn to control not only the actions of artificial intelligence, but also their own actions in relation to it. Artificial intelligence does not necessarily need to rebel against humanity; it is enough to convince humans that they can win by destroying each other.

www.bankingnews.gr

Ρoή Ειδήσεων

Σχόλια αναγνωστών

Δείτε επίσης